Abstract

We present the emerging science of Mind Genomics, to understand people’s responses to health-related issues, specifically pain. Mind Genomics emerge out of short, affordable, scalable, east-to-run experiments. The topic, here pain, is deconstructed into four questions, each with four separate answers (elements.) The answers are combined into vignettes, presented to respondents, who rate the entire vignette. Emerging from the study are the ratings and the response times to the vignettes, both of which are deconstructed into the contributions of the different underlying elements which the vignettes comprise. The answers cannot be gamed, and the data quickly reveal what is important to the individual, as well as revealing the existence of new-to-the-world mind-sets which differ in the pattern of elements that they find important. Mind Genomics  provides the opportunity to understand the person’s needs and wants for specific health as well as other experiential situations where human judgment is relevant.

Introduction

Pain is an inevitable companion in our life’s journey. Pain is defined through its association with actual or potential tissue damage, denoting it as a necessary characteristic of the experience, but also recognizing that events other than tissue damage can serve as determinants, consistent with a bio psychosocial model of pain [1,2]. This definition of pain denotes multiple causal factors underlying pain, beyond the issue pathology.

There is no dearth of studies on pain, whether these studies are report of pain from one’s everyday life [3], a topic dealt with in medicine [4], and a topic of scientific investigation [5]. When we talk about pain, can we probe into the mind of the person beyond simply the report, beyond a simplistic scale? Can we move beyond simple indicators, approaching a more detailed description of one’s pain but yet not forcing the respondent to become a scientist?

Pain, a highly subjective phenomenon, often refers to a sensory experience resulting from actual damage to the body or from non-bodily damage [6]. Pain may be influenced by psychological mechanisms such as: attention, emotion, beliefs and expectations [7].

In general, there are two different classifications of physical pain, visceral and somatic. Visceral pain originates in the internal organs whereas somatic pain stems from skin, muscle, soft tissue, and bone. There are many types of pain which fall under these categories. A person’s pain can also be classified as acute or chronic. Pain can be described as nerve pain, psychogenic pain, muscle pain, abdominal pain, back pain, pelvic pain, etc.

Subjective pain is influenced by its intensity and by interventions to treat the pain. Expectations and attitudes towards pain, may stem from psychological processes that are fundamental to learning across various sensory experiences and affect. Understanding expectations and attitudes towards pain may help us form communication messaging to help individuals deal more effectively with their chronic pain.

The subjective nature of pain makes it difficult to test the actual nature of perceived pain across populations, within a country, and in different countries. There are accepted methods of testing the actual perception of pain, specifically pain thresholds and pain tolerance, as well as psychophysical scaling of pain. One example is measuring the time one can submerge a limb in an ice bath, to test the ability of subjects to tolerate pain under varying conditions, most notable with the testing of analgesics of anesthetics. These methods give a measure of the all-or-none response to pain, and even the qualitative nature of the pain, but do not give a sense of the mind of the person who is undergoing the pain.

Increase in pain accompanies one’s beliefs that a certain treatment will cause pain or increase one’s symptoms overtime [7]. Negative beliefs regarding pain and its effects may occur in some types of chronic pains. To test whether expectations affect pain, studies tested the extent to which expectations influenced physiological responses among individuals. Placebo treatments truly reduced pain intensity [8–12]. These studies also indicated that short-term expectations varied and strongly affected perceptions of pain and pain-evoked responses [13].

Other studies linked differences in expectations regarding pain to the magnitude of responses to pain treatments [14]. Research on the relationship between expectations and pain experiences, showed that expectations about treatments and painful stimuli profoundly influenced behavioral markers of pain perception [7].

Pain treatments also bring positive changes in negative emotions [15]. Expectations affect pain through attention, executive functioning, value learning, anxiety and negative emotions [16]. Attitudes towards pain such as anxiety raised subjective pain. Pain is, thus a complex experience, involving sensory, motivational, and cognitive components. Affect any one of these components may change one’s attitudes towards pain [7].

Whereas studies indicate that beliefs influenced pain experience, it is unclear to what extent psychological processes such as attention, anxiety and emotions affect choice of treatments and what communication messages may mediate the effects of these psychological processes. This study tests communication messaging that affect emotion, attitudes towards pain and choice of treatment for pain.

In his book, Pain: The Gift Nobody Wants, author Paul Brand, MD describes his observations across cultures. Growing up as the child of missionaries in India and then moving to the US, Brand noted the difference in pain and suffering that existed in the East versus the West. He noted that, “as a society gained the ability to limit suffering, it lost the ability to cope with what suffering remains”. He stated that he believed that Easterners have learned to control pain at the level of the mind and spirit whereas, Westerners tend to view pain and suffering as an injustice or failure and an infringement on their right to happiness [17].

In the newly developing science of Mind Genomics we attempt to demonstrate a richer understanding of one’s inner life by presenting the respondent (or ill/healthy pain sufferer, here) with vignettes describing the inner experience, instructing the respondent to rate the fit of the vignettes, one at a time, and then estimating the degree to which each of the elements of the vignette ‘fits’ the respondent.

Method

Mind Genomics as an emerging science has been previously presented [18]. Mind Genomics works by presenting respondents with vignettes, combinations of statements which together tell a story. The respondent is instructed to judge the vignette, rating the vignette as a totality. The rating scale for this study is simply ‘How well does this describe you?’

The statements, elements in the language of Mind Genomics, present simple ideas. The approach requires the construction of four questions which ‘tell a story.’ For each question, the researcher is required to provide four answers, all expressed in simple language.  Table 1 presents the four questions, and the four answers to each question. Ideally, the questions and answers should deal with the topic, here pain, but need not mention pain directly. Rather, the questions and answers should be relevant to the topic.

Table 1. The four questions and the four answers to each question.

The answers in Table 1 are combined by experimental design into a set of 24 vignettes, with each vignette comprising 2–4 elements. Table 2 shows an example of the first six vignettes. The elements appear an equal number of times. Each of the 16 elements is, by design, statistically independent of every other element.

Table 2. The first seven vignettes for the first respondent, created by the experimental design. The table shows the combinations, then the combinations transformed into binary, and then the ratings.

Each respondent evaluates a unique set of 24 vignettes. The underlying mathematical structure of the experimental design is maintained, but the specific combinations are changed, in a permutation scheme which preserves the mathematical properties of the design [19]. The permutation covers many more combinations of elements compared to the standard approach of creating one experimental design and presenting that design to many respondents.  The Mind Genomics achieves stability by testing many combinations, each a single time, but the expanded coverage ensures that a great of the ‘space of combinations’ is covered. It is difficult to be very ‘wrong’ with a Mind Genomics study because the scope. In contrast, traditional research works with a very small experimental design, e.g., equivalent to the combinations tested by one person, but the combinations are tested by many respondents in order to obtain a stable estimate of the value for each combination.

Mind Genomics and traditional statistics are on opposite sides in terms of what generates valid data. Is valid data obtained by sampling a few of the many possible combinations, albeit with stability for each point (traditional), or by sampling a great many of the combinations, albeit with less stability at any point. A good analogy to Mind Genomics is, metaphorically, the MRI, which discovers the configuration of tissue by taking different ‘snapshots’ and integrating them into one picture.  With the permuted experimental one need not ‘be sure’ that the limited number of combinations is the correct set to represent the total set of possible alternatives. With as few as 25 respondents, the number of respondents participating, generating a total of 720 different combinations has covered the space quite well.

Running the Mind Genomics experiment

The experiment is run on the web, typically with respondents from a specific population who have agreed to participate (e.g., those being treated for a condition), or more typically with respondents recruited from the general population, when the objective is a quick ‘scan’ of what is important.  The base sizes of these studies range from 25 for an exploration to 500 for a massive deconstruction of the population into different mind-sets.  The more typical base size of 25–50 respondents reveals quite a bit about the nature of people’s minds with regard to a particular issue.  This study shows the type of learning emerging from this small base size of respondents from the general population, and can be followed with many different studies to follow up on various interesting aspects.

The elements, answers to the questions, are created by experimental design [20]. The 16 elements are combined into 24 combinations or vignettes, similar in structure to the vignettes shown schematically in Table 2. The vignette can be presented on smartphones, tablets, or PC’s.

Although the respondent might feel that the vignettes are created in a random fashion, the reality is just the opposite. The vignettes are created within the framework of the design, which prescribe the exact combinations. The elements are placed one atop the other, centered, without any connectives, making the respondent’s task easier as the respondent ‘grazes for information’.

The experimental design ensures that the elements are statistically independent; appear several times against different backgrounds provided by the other elements in the vignette. Each respondent evaluates a unique set of 24 vignettes, permuted as noted above, so that the design structure is maintained but the specific combinations are new. The permutation system allows a great deal of the design space, or combinations, to be tested, and allows the information to emerge even when the researcher has absolutely no idea what will be important and what won’t. In other words, Mind Genomics is a discovery system, and not a confirmation system. One can learn quickly from a base of zero knowledge, simply by doing 1–4 easy studies of different facets of a topic.

The respondents who participated were US residents, members of a 10+ million world-wide panel of Luc.id Inc., who had previously agreed to participate in these studies for a reward administered by the panel provider. All respondents participated anonymously. The only information about the respondent was age, gender, and the answer to the third question about what type of pain they had.  There were five answers to the third question, three dealing with chronic pain of various sorts, and two saying either ‘no pain,’ or ‘not applicable.’  All respondents were classified by gender, age, and by either pain/yes versus pain/no.

Preparing the data for analysis

The respondent assigns a rating to assess ‘How much does this describe how you feel’. The low anchor, 1, is ‘not at all.’ The high anchor, 9, is ‘very much.’ The Mind Genomics program bifurcates the scale, dividing it into the lower part, ratings of 1–6, transformed to 0, plus a very small random number (<10^–5), and a high part, ratings of 7–9, transformed to 100, plus a very small random number. The bifurcation comes from the decades of experience which suggest that managers and scientists alike do not ‘understand’ the meaning or use of the Likert or category scale, but they easily understand the meaning of a no/yes, binary scale.  The choice of where to bifurcate is left to the researcher. Thirty-five years of experiments suggest that a 2/3 vs 1/3 division seems to work well.  The small random number added to the binary transformed data ensures that when it is time to run the OLS (ordinary least-squares) regression on the data at the level of the individual respondent, there will not be a ‘crash’ of the regression program when the respondent confined the ratings to either the low range (1–6) or to the high range (7–9.) Either of those two cases produces all 0’s or all 1’s, crashing the regression. The small random number ensures that there is variability in the dependent variable, the binary transformed data.

How the different elements drive the binary transformed rating

Table 3 shows the parameters and relevant statistics for the additive model created from the ratings of the total panel, after transformation to a binary scale. The model itself is a simple linear equation of the form: Binary Rating = k₀ + k₁(A1) + k₂(A2) … K₁₆(D4). The experimental design allows us to create the model either at the level of the individual respondent or at the grand level, combining all of the data from the ‘relevant’ respondents, with relevant being

Table 3. Parameters of the model for ‘Fits Me’ after binary transformation. The data come from the Total Panel (720 observations, 24 tested vignettes from each of 30 respondents.) The table is sorted in descending order of coefficient for ‘describes me.’ At the right is the associated coefficient for response time.

The analysis suggests the following:

1. Additive constant, the expected binary value in the absence of elements: Without any elements, the likely response that the vignette will ‘describe me’ is about 46%. By design, all vignettes comprised 2–4 elements, so the additive constant is an estimated parameter.  Thus, the value of 46 for additive constant says that half the time respondents will answer that whatever appears will describe them. It is the elements which must do the work to move beyond this almost 50% agreement rate. It is worthwhile commenting here that this baseline of 46% is modest. When the topic is credit cards and the rating is ‘interested in acquiring this credit card,’ the additive constant plummets to about 10–15. When the topic is pizza and the rating is ‘interested in eating this pizza,’ the additive constant skyrockets to 60–70.
2. There are no very strong elements for the total panel: That is, no element drives the description of ‘me.’ This weakness can either be the result of choosing the wrong elements, or the result of dealing with two or perhaps even three or more different populations, who describe their impressions by different terms, and who may live in quite different worlds of pain.
3. The highest scoring element is C2, I would like exercises and stretches that reduce pain. This element generates a coefficient of only 6, and has a t-statistic of 0.95, with a probability of 0.34 that it came from a distribution with a true mean of 0. That is, it’s quite likely that were we to do this study again, we would come up with a coefficient much lower than 6, probably 0 or thereabouts.
4. The remaining elements do not ‘fit’ the respondent:  It may well be that the elements are simply incorrect and others will fit the respondent better, or more likely that we are dealing with a segmented population of individuals, some of whom feel that an element ‘fits them,’ whereas others feel that the same element ‘does not fit them.’ In such a situation the responses cancel each other, and we are left with a coefficient around 0, denoting ‘no fit.’

Key subgroups

We know three additional things about the respondent based upon the self-profiling questions completed during the study. The first is gender, the second is age, and the third is whether or not they suffer pain on a regular basis. In this computerized application, the respondent is required to select one of two genders (male/female), and required to put in the year of birth, which provides age.  The third question is left to the discretion of the researcher. In this study is the selection of pain, with five options. Two options are defined as ‘no pain’ (actual selection of ‘no pain’ as an answer, selection of not applicable). The remaining three options as pain (i.e. pain in the limbs, back, etc.).  We will look at gender, age, and self-reported pain as the three self-defined subgroups. We will also explore two new subgroups, mind-sets inherent in the population but revealed by understanding patterns of responses, behavioral patterns, rather than self-classification.

The focus of interest in Mind Genomics studies is on the additive constant as the ‘baseline,’ and then on the ‘story’ told by the winning elements.  These elements are operationally defined as having a value of +6.51 or higher, which becomes 7 when rounded to the nearest whole number.

Gender

1. Males show a higher additive constant than do females (57 vs 38). In the absence of elements, men are more likely to say that a vignette ‘describes ME.’  Women are less likely to say that, and require more specification.
2. We get a good sense of what is important by looking at the elements which are most positive (most like me), and most negative (least like me)
3. For men, the single phrase which most describes them is

   C2: I would like exercises and stretches that reduce pain

4. For men, the single phrase which least describes them is

   C1: I would like to have a diet that is tailored to reduce my pain

5. For women, the two phrases phrase which most describe them are

   B2: I try to deal with the pain to work through it,

   A1: Pain bothers me all over my body. The degree of fit is less, however, for these elements than the corresponding best fits for males.

6. For women, the phrase which least describes them is

   B4: I just like taking a pill that deals with the pain.

Age: Under 50 versus 50+

Respondents provided the year of their birth. One respondent did not provide the year and was eliminated from this particular analysis by age.

1. Surprisingly, the additive constant is much higher for the younger respondents versus the for the older respondents (48 vs 31.)
2. For the younger respondents, there are no strong elements which fit them. The two elements which most describe them are those which suggest control over the pain:

   C2: I would like exercises and stretches that reduce pain

   D3: The doctor should set me up with a system for me to follow

3. For the younger respondents, the two elements which least describe them are those which suggest passivity, and no control over the pain.

   B1: The doctor explains to me how to deal with the pain

   B4: I just like taking a pill that deals with the pain.

4. For the older respondents, the two elements which most describe them are actual experience to reduce the pain, as well as a description of the experience.

   A3: The pain radiates and makes it difficult to function

   C2: I would like exercises and stretches that reduce pain

5. For the older respondents, the three elements which least describe them is passivity

   D1: The doctor should give me advice

   C4: I would like a prescription that gives me the medication I need to feel better

   C1: I would like to have a diet that is tailored to reduce my pain

No pain versus pain

As part of the self-profiling classification, the respondents selected the type of pain, if any, afflicting them. The respondents who check any of the three types of pain assigned to the group saying YES. The remaining respondents were assigned to the group saying NO.

1. The additive constant is virtually the same, 46 vs 48, meaning that in the absence of elements in the vignette; a little fewer than 50% of the responses will be ‘describes me.’
2. For those with pain, the phrase which most describes them is

   C2:  I would like exercises and stretches that reduce pain.

3. For those with pain, the element which least describes

   C1:  I would like to have a diet that is tailored to reduce my pain

4. For those with no pain, virtually no element most describes them
5. For those with no pain, many elements least describe. The strong element which least describes is

   C4: I would like a prescription that gives me the medication I need to feel better

Mind-Sets: Dividing respondents by the patterns of their coefficients for a specific topic

We have just seen that there are some differences in terms of ‘describes me’ across genders, and across those who define themselves as having pain versus no pain. These are ways that people describe themselves. People may differ in ways that the researcher cannot describe in simple terms, or even in way that they themselves don’t understand.

A major tenet of Mind Genomics is that within any topic area, such as the description of pain presented here, there are fundamental differences across people, differences that are obvious once demonstrated, but differences limited to a single topic area.  This is the case of the data here. Even within the small sample of 30 respondents we can extract two, possibly three different mind-sets. The method for extracting mind-sets has been previously described [21]. Quite simply, the technique is a matter of clustering the respondents into two or three groups based upon the pattern of their 16 coefficients. The statistical method of clustering is well accepted [22] All that remains is the clustering, extracting the small groups with the property that these mutually exclusive groups represent different ways of thinking about the topic.

Table 4 shows the results for the two mind-set segments emerging from the clustering of the 30 respondents. A base size of 25–30 suffices to reveal the nature of these different mind-sets, especially because the segments are so obviously different and interpretable.

Table 4. Coefficients for the binary-transformed scale ‘Describes me’ across gender, age, pain, and mind-set, respectively. Coefficients of +7 or more are presented in bold, and shaded.

1. Mind-Set 1 (wants a cure) begins with a low additive constant, 37. To them, it’s not the general response which ‘describes me’ but rather the specific phrase. Mind-Set 1 suffers pain, and wants a cure. Here are the elements which Mind-Set 1 feels best describes them:

   A1: Pain bothers me all over my body

   A3: The pain radiates and makes it difficult to function

   C2: I would like exercises and stretches that reduce the pain

2. Mind-Set 1 do not want simple medical treatment which will alleviate their pain. Here is the element which is they feel least describes them:

   B4: I just like taking a pill that deals with the pain.

3. Mind Set 2 (simplicity through the doctor) shows a higher additive constant, 54. Mind-Set 2 is less discriminating among elements. Mind-Set 2 wants simplicity. Here is the one element that they feel best describes them:

   D3: The doctor should set me up with a system for me to follow

4. Mind Set 2 does not want to take responsibility. Here are the elements that they feel least describe them:

   C1: I would like to have a diet that is tailored to reduce my pain

   B1: The doctor explains to me how to deal with the pain

   A3: The pain radiates and makes it difficult to function

Response times as a measure of cognitive processing of information

At the same time that the respondents were reading the vignettes, the response time was being measured. Response time is operationally defined as the time between the appearance of the vignette and the assignment of the rating. The experiment was executed on the internet.

 The respondent was unaware of response time being measured, being instructed simply read the vignette and assign a ‘gut-level’ judgment. Occasionally, in about 10% of the cases, the response time was longer than 10 seconds, suggesting that the respondent was doing something as well, so-called multi-tasking. Those response times of 10 seconds or longer were recoded as 10 seconds. Figure 1 shows the distribution of the 720 response times (30 respondents, each evaluating 24 vignettes)

Figure 1. Distribution of response times for the total panel of 30 respondents, each rating 24 unique vignettes.

Response time patterns for different subgroups

The measurement of response times as a key feature of Mind Genomics began during the summer of 2019. In the studies run since that introduction, the response time data suggests that when the topic deals with an important health issue, the respondents spend a long time reading the vignette, and thus their response times are long, often 1.0 seconds or longer. When the topic deals with something commercial or ‘fun’ the response times are very short, around 0.2 – 0.7 seconds.

Table 5 presents the response time coefficients for the key subgroups. The model for response time is written in the same way as the model for the binary transformed rating, with the key difference being that that the model for response time does not have an additive constant. The ingoing assumption is that the response time is 0 when there are no elements in the vignette.

Table 5. The coefficients for the response time models. The models do not feature an additive constant.

In Table, coefficients of 2.0 or higher are shaded and shown in bold. These are the elements to which the respondent pays attention.  There are some simple patterns which emerge from visual inspection of these elements that are processed ‘more slowly.’

1. For gender, males focus on the description of symptoms.

   A4  The pain is minor but frequent and annoying

   B2   I try to deal with the pain to work through it

   B3   I’m happy when I can use a device that delivers therapeutic solution

   C4  I would like a prescription that gives me the medication I need to feel better

   B4   I just like taking a pill that deals with the pain.

2. For gender, females want a relationship, or at least someone/something external to them.

   D3  The doctor should set me up with a system for me to follow

   B3   I’m happy when I can use a device that delivers therapeutic solution

   D2  The doctor should give me a shot that delivers long term relief

3. For age, those under 50 focus on only one element:

   D3  The doctor should set me up with a system for me to follow

4. For age, those 50+ focus on a number of phrases, most dealing with methods to assure pain reduction

   B3   I’m happy when I can use a device that delivers therapeutic solution

   B1   The doctor explains to me how to deal with the pain

   A4  The pain is minor but frequent and annoying

   B4   I just like taking a pill that deals with the pain.

   D3  The doctor should set me up with a system for me to follow

   B2   I try to deal with the pain to work through it

   D4  The doctor should give me a regular schedule of visits to treat my pain

   D2  The doctor should give me a shot that delivers long term relief

5. For pain, those with PAIN YES, i.e., who say they suffer from one or another pain, the focus is on what stops the pain, i.e., assure pain reduction

   B3   I ‘m happy when I can use a device that delivers therapeutic solution

   B2   I try to deal with the pain to work through it

   D3  The doctor should set me up with a system for me to follow

   C3  I would like regular therapy sessions to reduce my pain

   C4  I would like a prescription that gives me the medication I need to feel better

6. For pain, those with PAIN NO, i.e., who say that they do not suffer from pain, the focus is on descriptions of pain

   A4  The pain is minor but frequent and annoying

   D3  The doctor should set me up with a system for me to follow

   A1  Pain bothers me all over my body

7. For Mind-Sets, Mind-Set 1 (Wants a cure)

   D3  The doctor should set me up with a system for me to follow

   B2   I try to deal with the pain to work through it

   B3   I’m happy when I can use a device that delivers therapeutic solution

8. For Mind-Sets, Mind-Set 2 (Simplicity through the doctor)

   A4  The pain is minor but frequent and annoying

   B3   I’m happy when I can use a device that delivers therapeutic solution

   D4  The doctor should give me a regular schedule of visits to treat my pain

Finding the mind-sets in the population using a PVI (Personal Viewpoint Identifier)

The mind-sets reveal different ways of perceiving the nature of pain.  The mind-sets represent a way to divide what is likely a continuum of feelings and points of view into at least two distinct groups, a division which may provide further understanding, and certain a division that can be used to deal with patients in different, and possibly more appropriate fashion.

Table 6 shows, however, that it’s unlikely to identify mind-sets by their age and gender. It is also quite possible that there are no direct classifications of who a person ‘is’ or what a person ‘experiences’ which can easily assign a person to one of these two mind-sets.

Table 6. How the two emergent mind-sets for pain distribute on the self-profiling classification in terms of age, sex, and experience of pain.

An alternative way to assign new individuals to mind-set has been developed by author Gere. It is called the PVI, the personal viewpoint identifier. The PVI comprises a set of six questions, answered with one of two answers, no or yes.  The pattern of the answers to the six questions assigns the respondent to one of the two mind-sets.  Figure 2 shows the PVI questionnaire at the left, and the response emerging, given either to the physician and/or to the patient/client.  The questions themselves are taken from the actual study. These are the answers or elements, now turned into questions.

The PVI can be deployed along with additional information obtained during the questions. Thus, Figure 2 shows that the respondent, a new person not part of the previous study establishing the PVI, is asked for his or her email. Other questions can be asked, to relate mind-set membership to external variables, whether of a medical/health nature, or of a life-style nature.

Discussion & Conclusions

Since pain is a complex sensation involving sensory, motivational, and cognitive components, and affecting any one of these may change one’s attitudes towards pain [7]; we tested the effect of communication messaging, across mind-set segments towards pain. We tested how each min-set segment we identified emotionally responds to chronic pain, and which treatment choices are preferred by attitudinal mind-sets towards pain.

People who belong to the first mind-set segment feel the pain as radiating and challenging their daily functioning. The pain is very bothersome, but they choose to alleviate it by exercises and stretching. They chose to avoid medical treatment to simply deal with the pain and its ramifications.  People belonging to the second mind-set segment also view their chronic pain as radiating and challenging their daily functioning.  They, however, choose to simply take pain medication their doctor will prescribe.  They expect their doctor to also set them up with a system to follow.  In addition, they do not want to take responsibility for self-managing the illness which causes their pain. They prefer to avoid a diet that is tailored to reduce their pain.

Figure 2. The PVI created for the pain study. The link for the PVI as of this writing (Feb. 2019) is: http://162.243.165.37:3838/TT13/

This study also illustrated how a medical professional may easily identify the mind-set segment to which a patient belongs and accord communication messaging to patient choices and values. Identification of the mind-set to which a patient belongs may assist in building patient-physician trust resulting in higher patient adherence and better implementation of patient-centered care [21].

Mind Genomics provides the ability to segment out populations that share a common mind type and thereby help identify the possibility of determining the types of pain that a person is most likely to experience. It may help answer the question of why people with the same disease experience pain in profoundly different ways. By mind-typing patients who share ailments, Mind Genomics may aid in helping tailor a treatment plan best suited to that individual lying within a disease spectrum.

In light of the current opioid epidemic, it more important, now more than ever, to address how to customize pain treatments to individuals. There are many modalities to treat pain. In the West, pain medications are the first line of treatment. These medications include narcotics/opiates, Non-Steroidal Anti-Inflammatory Drugs (NSAIDs), acetaminophen, certain antidepressants, muscle relaxants, anticonvulsants, corticosteroids, local anesthetics, and most recently medical marijuana. Other modalities such as Transcutaneous Nerve Stimulation (TENS), implantable spinal cord stimulators, meditation and biofeedback are also used to help combat pain. Health care professionals who specialize in pain management use experience and training to try and help tailor treatment regimens to the individual patient. But a tool like Mind Genomics may help the practitioner go beyond the current protocols and prejudices of current practice. Mind Genomics may provide a “cheat sheet” to the patient’s mind and help provide a short cut to success by focusing on pathways that will more likely work for a given patient and eliminating the pathways that will waste time and resources.

References

1. Hadjistavropoulos T, Craig KD, Duck S, Cano A, Goubert, L, et al. (2011) A biopsychosocial formulation of pain communication. Psychological bulletin 137: 9.
2. Williams AC, Craig KD (2016) Updating the definition of pain. Pain 157: 2420– 2423. [crossref]
3. Baker KS, Gibson S, Georgiou-Karistianis N, Roth RM, Giummarra MJ (2016) Everyday executive functioning in chronic pain: specific deficits in working memory and emotion control, predicted by mood, medications, and pain interference. The Clinical Journal of Pain 32: 673–680.
4. Morrison RS, Maroney-Galin C, Kralovec PD, Meier (2005) The growth of palliative care programs in United States hospitals. Journal of Palliative Medicine 8: 1127–1134.
5. Schug SA, Palmer GM, Scott DA, Halliwell R, Trinca J (2016) Acute pain management: scientific evidence, 2015. Medical Journal of Australia 204: 315–317.
6. Loeser JD, Treede RD (2008) The Kyoto protocol of IASP Basic Pain Terminology. Pain 137: 473–477. [crossref]
7. Atlas LY, Wager TD (2012) How expectations shape pain. Neurosci Lett 520:140–148. [crossref]
8. Goffaux P, de Souza JB, Potvin S, Marchand S (2009) Pain relief through expectation supersedes descending inhibitory deficits in fibromyalgia patients. Pain 145: 18–23.
9. Goffaux P, Redmond WJ, Rainville P, Marchand S (2007) Descending analgesia–when the spine echoes what the brain expects. Pain 130: 137–143. [crossref]
10. Matre D, Casey KL, Knardahl S (2006) Placebo-induced changes in spinal cord pain processing. Journal of Neuroscience 26: 559–563.
11. Price DD, Craggs J, Verne GN, Perlstein WM, Robinson ME (2007) Placebo analgesia is accompanied by large reductions in pain-related brain activity in irritable bowel syndrome patients. Pain 127: 63–72.
12. Alkes L. Price, Arti Tandon, Nick Patterson, Kathleen C. Barnes, Nicholas Rafaels, et al (2009) Sensitive Detection of Chromosomal Segments of Distinct Ancestry in Admixed Populations. PLoS Genet 5: 1000519.
13. Atlas LY, Bolger N, Lindquist MA, Wager TD (2010) Brain mediators of predictive cue effects on perceived pain. J Neurosci 30: 12964–12977. [crossref]
14. Watson A, El-Deredy W, Iannetti GD, Lloyd D, Tracey I, et al. (2009) Placebo conditioning and placebo analgesia modulate a common brain network during pain anticipation and perception. PAIN 145: 24–30.
15. Wager TD, Atlas LY, Leotti LA, Rilling JK (2011) Predicting individual differences in placebo analgesia: contributions of brain activity during anticipation and pain experience. Journal of Neuroscience 31: 439–452.
16. Flaten MA, Aslaksen PM, Lyby PS, Bjørkedal E (2011) The relation of emotions to placebo responses. Philosophical Transactions of the Royal Society B: Biological Sciences 366: 1818–1827.
17. Brand PW, Yancey P (1993) Pain: the gift nobody wants. New York, HarperCollins Publishers.
18. Moskowitz HR, Gofman A, Beckley J, Ashman H (2006) Founding a new science: Mind genomics. Journal of sensory studies 21: 266–307.
19. Gofman A, Moskowitz HR (2010) Isomorphic permuted experimental designs and their application in conjoint analysis. Journal of Sensory Studies 25: 127–145.
20. Box GE, Hunter JS, Hunter WG (2005) Statistics for experimenters: design, innovation, and discovery (Vol. 2). New York: Wiley-Interscience.
21. Gabay G, Moskowitz HR, Silcher M, Galanter E (2017) The New Novum Organum: Policies, Perceptions and Emotions in Health. Pardes-Ann Harbor Publishing.
22. Moskowitz HR, Martin DM (1993) How computer aided design and presentation of concepts speeds up the product development process. Paper presented at the ESOMAR Congress, September, 1993, Copenhagen.
23. Eippert F, Finsterbusch J, Bingel U, Büchel C (2009) Direct evidence for spinal cord involvement in placebo analgesia. Science 326: 404. [crossref]
24. Moskowitz, HR (2012) ‘Mind genomics’: The experimental, inductive science of the ordinary, and its application to aspects of food and feeding. Physiology & behavior 107: 606–613.

Abstract

Background: Effective hospital teams can improve outcomes, yet, traditional hospital staffing, leadership, and rounding practices discourage effective teamwork and communication. Under the Accountable Care Unit model, physicians are assigned to units, team members conduct daily structured interdisciplinary bedside rounds, and physicians and nurses are jointly responsible for unit outcomes.

Objectives: To evaluate the impact of ACUs on patient outcomes.

Design: Retrospective, pre-post design with concurrent controls.

Patients: 23,406 patients admitted to ACU and non-ACU medical wards at a large academic medical center between January 1, 2008 and December 31, 2012.

Measures: In-hospital mortality and discharge to hospice, length of stay, 30-day readmission.

Results: Patients admitted to ACUs were less likely to be discharged dead or to hospice (-1.8 percentage point decline [95% CI: -3.3, -0.3; p = .015]) ACUs did not reduce 30 day readmission rates or have a significant effect on length-of-stay.

Conclusions: Results suggest ACUs improved patient outcomes. However, it is difficult to identify the impact of ACUs against a backdrop of low inpatient mortality and the development of a hospice unit during the study period.

Keywords

quality improvement, teamwork, hospital medicine, care standardization

Introduction

Under the traditional model of inpatient staffing, hospitals nurses and allied health professionals are assigned to a unit, while hospital medicine physicians treat patients on multiple units. Care is delivered asynchronously. Physicians see patients when their schedules permit, usually early in the morning or in the late afternoon and update orders at those times. Nurses and other professionals care for patients separately. They may not see the physician during rounds, and their priorities for patient care may be different from those of the physician. In our experience, they often obtain information from second-hand sources or the often difficult-to-decipher notes in patients’ charts.

The traditional, physician-centric model of inpatient care poses significant coordination and incentive problems. Beginning in October 2010, Emory University Hospital re-organized two medical units into Accountable Care Units (ACU® units). In the ACU care model, hospital-based physicians are assigned to a home unit where they can focus on the patients in the unit and work with the same nurse team. By assigning physicians to home units with other unit-based personnel such as nurses and having teams engage in structured interdisciplinary bedside rounds, ACUs enable clinicians to recognize preventable hospital complications and signs of deterioration or diagnostic error that might otherwise have been missed and implement a coordinated response.

Previous publications on the ACU model have been mostly descriptive in nature [1–4]. Using electronic medical records and a pre-post study design with concurrent controls, we retrospectively evaluated the effect of ACUs on patient mortality, length of stay, and readmissions at Emory University Hospital.

Methods

Intervention

Emory University Hospital is a 500 bed teaching hospital in Atlanta, Georgia. Prior to the implementation of ACUs, hospital medicine physicians at Emory University Hospital treated patients in as many as eight units. In the first unit to be organized into an ACU, patients were divided between five physician care teams prior the re-organization. Beginning in October 2010, Emory University Hospital assigned two physician care teams to each of two newly-constituted ACU units. ACUs combine a number of interventions, some of which have been implemented at other hospitals [5–8] , into a single, cohesive bundle.

ACU physician teams were assigned to units and included one hospital medicine attending physician, one internal medicine resident, and three interns. Within an ACU, two teams rotated call schedules over a 24 hour period. The team on-call admitted every patient who arrived at the unit. The same nurse teams continued to staff each unit as before the reorganization.

ACUs standardize communication through a series of brief but highly scripted intra- and inter-professional exchanges to review patients’ conditions and care plans. Each shift change begins with a five minute huddle where the departing staff hands over the unit to the incoming staff. During the huddle, the departing staff alerts the incoming staff to patient- and quality-related issues. After the huddle, nurses hand over individual patients at the bedside using a structured format, highlighting patient-level factors that might indicate patient instability or are outside the expected range. Once a day, each patient’s care team meets for structured interdisciplinary bedside rounds. Structured interdisciplinary bedside rounds bring the bedside nurse, attending physician, and unit-based allied health professionals to the bedside every day with the patient and family members to review the patient’s current condition, response to treatment, care plan, and discharge plan collaboratively [5–8]. Evidence-based actions, such as “bundles” to prevent hospital acquired conditions, are embedded in structured interdisciplinary bedside rounds, and reported on by the patient’s nurse. A scripted, standard communication protocol reduces extraneous communication and focuses the structured interdisciplinary bedside round team’s attention on aspects of patients’ conditions that are responsive to staff attention and effort.

A unit leadership dyad, consisting of a nurse manager and senior physician, set explicit expectations for staff and manage unit process and performance. Physicians operating in the traditional model may be unaware of unit-level quality protocols and outcome measures. As part of the re-organization, a data analyst prepared quarterly unit-level performance reports describing rates of in-hospital mortality, blood stream infections, 30-day readmissions and patient satisfaction scores and length of stay. These reports are used by hospital administrators to set goals for the ACU leadership team and may figure into the performance evaluations of ACU administrators. Readers interested in additional details about the ACU model are urged to consult previous publications [1–4].

Following implementation of ACUs, physician teams assigned to ACUs saw patients on only 1.5 units, with 90% of their patients located in the ACUs, compared to non-ACU physician teams, which cared for patients spread across 6 to 8 units every day.¹ The number of patient encounters per day for the ACU physician teams increased from 11.8 in the year before the ACUs (when the teams were not unit based) to 12.9 in the four years following implementation [1]. No changes were made to nurse staffing levels (1 to 4 or 5 nurses per patient).

During the study period, Emory University Hospital created two ACUs, but medical patients were also admitted to seven other units in the hospital. The units that became ACUs were selected because nearly all the patients were under the care of hospital medicine attending physicians so we could designate them as hospital medicine units. In other units, hospital medicine patients were mixed in with patients from other specialties (for example, cardiology). The assignment of patients to ACUs or other medical units was determined by bed control officers based on a mix of criteria that can include bed availability, relative patient wait times, and individual judgement. Bed managers know patients’ names, medical record number, and admitting diagnosis when they assign patients to units. They do not know have access to other prognostic indicators.

Study Sample

The study sample includes patients ages 18 and older admitted to the medical units of Emory University Hospital between January 1, 2008 and December 31, 2013. Following an intent-to-treat framework, we grouped patients who were transferred into ACUs during their hospital stay with non-ACU patients. Patients admitted to surgical, orthopedic, observation, or other specialty units (e.g. medical oncology) were excluded from the analysis, as were patients with cystic fibrosis who are treated only within one of the two ACUs. Patients in the control group were spread across 38 units, though 70% were in just 8 of these units.

Data and Outcome Variables

All study variables are captured in Emory’s internal electronic medical record and administrative data systems. We evaluated the impact of ACUs on in-hospital mortality, discharge to hospice, length of stay, readmission or emergency department visit to Emory University hospital within 30 days, and hospital-acquired urinary tract infection and deep vein thrombosis and pulmonary embolism. We counted a patient as having hospital-acquired urinary tract infection and deep vein thrombosis and pulmonary embolism if their records listed ICD-9 codes for these condition but not if they were among the present-on-admission ICD-9 codes.

Emory University Hospital opened an on-site hospice during the study period in November 2010, potentially reducing the barriers to transferring patients from the hospital to hospice care. While discharge to hospice is in many cases an indication of appropriate care, the opening of the inpatient hospice complicates efforts to measure trends in in-patient mortality. The opening of the unit may be responsible for changes in the site of death for patients admitted to the hospital over time. For this reason, we highlight the impact of ACUs on the combined outcome of in-hospital death or discharge to hospice.

Statistical Analysis

We compared patient characteristics between ACUs and control units using chi-squared tests. We estimated the impact of ACUs on these outcomes using a difference-in-difference study design (equivalently, a pre/post study with a concurrent control group). The pre period was January 1, 2008 to October 31, 2010. The post period was November 1, 2010 to December 31, 2012. We calculated the change in outcomes between the pre and post periods among patients admitted to the units that became ACUs and the change among patients in the control group. The difference of these changes is the difference-in-difference estimator. It assesses changes in outcomes in the units that became ACUs relative to changes in the control group. It assumes that absent any change in policy (i.e., the implementation of ACUs), trends in outcomes among patients admitted to the ACUs would have mirrored trends among patients in the control group. We calculated 95% confidence intervals for unadjusted estimates using z-tests. We used logistic regression with robust standard errors to estimate adjusted effects for in-hospital mortality and hospice discharges and readmissions. We used Poisson regression with robust standard errors to estimate adjusted effects for length of stay. We calculated standard errors and 95% confidence intervals for the difference-in-difference estimator using the Delta method [9].

In multivariable analysis, we adjusted estimates for patient age group, sex, race, primary payer, admission source (hospital or skilled nursing facility versus other), and Elixhauser comorbidities (based on all diagnosis codes) [10] that were present in at least 2.5% of patients in the sample. About one-third of the sample had missing values for admission source. We included each Elixhauser comorbidities as a separate variable in the model rather than collapsing the conditions into a count to avoid imposing unnecessary restrictions on the relationship between conditions and outcomes. Conditions are not mutually exclusive.

Estimates from difference-in-difference models may be biased if there are pre-existing trends in outcomes that differ between ACU and non-ACU units. We tested for pre-existing trends by estimating a model that included, in addition to the variables described above, indicators for the years in the pre-period (2008 to 2010) and these year indicators interacted with treatment group (ACU versus non ACU). We assessed the significance of the year-group interactions and used a likelihood ratio test to compare the model fit with a model that omitted the year-group interactions [11].

Estimates of the impact of ACUs on in-hospital mortality and hospice discharge rates may be biased by differences in length of stay. An intervention that reduces length of stay but does not affect mortality rates will reduce in-hospital mortality rates by shifting the place of death from the hospital to the community. In a sensitivity analysis we assessed the robustness of logistic regression estimates by estimating a Weibull survival model with robust standard errors of the time to hospice discharge or in-hospital death. Records for patients who were not discharged to hospice or dead are censored.

Results

There were 23,403 patients included in the study sample, of whom 10,639 were admitted to the ACU units (including patients admitted to the units before they became ACUs) and 12,764 patients in the control group. There are significant differences in some of the characteristics of ACU and control group patients in the pre and post periods (Table 1), but most differences are qualitatively small. There are some clinically meaningful differences in patients’ diagnoses. For example, in the pre-ACU period, 8.2% of patients in the control group had a solid tumor compared to 6.7% in the ACU group.

The unadjusted proportion of ACU patients discharged to hospice or dead declined from 7.7% to 5.8% (Figure 1) or -2.0 (95% CI: -2.9, -1.0) percentage points. The unadjusted proportion of patients discharged to hospice and dead both declined. A reduction in in-hospital mortality rates accounted for 70% of the decline (= [2.5–1.1] ÷ 2).

Figure 1. Discharge destination in ACUs and control units

Table 1. Sample characteristics

The unadjusted proportion of patients in the control group discharged to hospice or dead declined from 7.9% to 7.1%, or -0.8 (95% CI: -1.7, 0.1) percentage points. A decline in the proportion of patients discharged dead was offset by an increase in the proportion discharged to hospice.

Adjusted estimates of the impact of ACUs are displayed in the last columns of Table 2. (Full regression results are available in the Appendix Table.) The adjusted estimate of the impact of ACUs on the composite outcome of discharged dead or to hospice is -1.8 (95% CI: -3.3, -0.3; p = .015) percentage points. The adjusted difference-in-difference estimate of the impact of ACUs on length of stay is negative but not statistically significant (-0.5 days [95% CI: -1.2, -0.3; p =.21]). The estimates for 30 day readmissions and hospital-acquired urinary tract infections are close to 0. The estimate of the impact of ACUs on the occurrence of pulmonary embolism/deep vein thrombosis was positive and borderline significant (0.6 percentage points [95% CI: -0.05, 1.3] p = .07).

Table 2. Changes in outcomes among ACU and non-ACU patients

Models that included year-group interactions rejected the hypothesis of pre-existing trends for discharge status and readmissions (see Appendix for details). In the survival model estimating time to in-hospital death or discharge to hospice, the hazard ratio for the interaction of the ACU group indicator and the post period indicator was less than one but did not achieve significance at α = 0.05 threshold (0.80 [95% CI: .63 to 1.00]; p = .052).

Discussion

Results indicate that ACUs reduced the proportion of patients discharged dead or to hospice. Length of stay declined in ACUs relative to control units, but the effect was mostly driven by an increase in length of stay in control units rather than a decrease in ACUs. ACUs did not appear to affect readmission rates. The opening of an inpatient hospice unit coincided with the introduction of ACUs, making it more difficult to identify the discrete impact of ACUs. However, physicians in all units of the hospital could transfer patients to the inpatient hospice unit, and so it should not have differentially affected outcomes in ACU versus non-ACU patients. The proportion of patients discharged to hospice actually declined slightly in the units that implemented ACUs. This pattern may reflect mean-reversion (the hospice discharge rate was higher in ACU units in the pre-period).

Given the low rates of in-hospital mortality in this patient population and hospital-wide efforts to reduce in-hospital mortality, patient discharge status may not be particularly sensitive to the quality of care. The regular rotation of residents and movement of other unit staff through the hospital may have spread some of the features of ACUs and their processes, resulting in hospital-wide improvements in outcomes.

Consistent with our predetermined analysis plan, we evaluated trends in ACU units relative to trends in control units. However, there were baseline differences in mortality rates and length of stay.

ACUs did not reduce the occurrence of hospital-acquired urinary tract infections and pulmonary embolism/deep vein thrombosis, at least as measured from billing records. It is unclear whether these results reflect a failure of ACUs to improve care or whether they reflect “surveillance bias” [12] : ACU teams may be more likely to recognize and diagnose patients with these conditions. The hospital implemented an initiative to more accurately document patients’ conditions during the study period, which may account for the increase in urinary tract infection rates.

Lacking access to information about patient health after discharge, we were unable to determine the impact of being admitted to an ACU on long-term outcomes. Patients discharged too early may experience adverse outcomes. We found that readmission rates were similar between the ACU and control groups, suggesting that patients were not being discharged from ACUs prematurely.

Although we evaluated the impact of ACUs in a single, large academic medical center, there are no elements or features of the ACU model that would prevent it from being expanded to other care settings. ACUs have already been implemented in community hospitals in the US, Canada (see http: //www.rqhealth.ca/department/patient-flow/accountable-care-unit accessed April 19th 2019) and Australia (see http: //www.cec.health.nsw.gov.au/quality-improvement/team-effectiveness/insafehands accessed April 19^th 2019).

Most prior studies on teams in inpatient and outpatient settings focus on single specialty teams (e.g. psychiatric care) and teams designed to address a specific quality issue (e.g., hospital acquired infections) [13,14].A recent report on the implementation of an Accountable Care Teams model, which shares many of the features of ACUs, at Indiana University Health Methodist Hospital found that implementation was associated with reductions in length of stay and costs but did not affect readmission rates or patient satisfaction [15].The assignment of hospitalists to units at Northwestern Memorial Hospital improved communication but did not increase physician-nurse agreement on patients’ care plans [16].

High risk industries with excellent safety records have recognized the value of teams to improving outcomes. ACUs, with their emphasis on patient-centered, interprofessional collaboration, were designed to address shortcomings of the traditional model of hospital organization. Our findings suggest that these and other features of the model were associated with reductions in the proportion of patients discharged dead or to hospice but did not affect other outcomes. Unfortunately, we were unable to assess the degree of fidelity of the study units to all features of the ACU model. Futures studies should include estimates of the extent to which units are implementing all four essential components of the model in estimating the effects of the model on distal outcomes.

Funding: Agency for Healthcare Research and Quality, R03 HS 022595-01

Conflicts of Interest: Dr Stein and Dr Chadwick are officers of 1Unit, a company that helps hospitals set up and run Accountable Care Units. Drs Howard, Shapiro, Murphy, and Ms Overton do not have any conflicts of interest.

References

1. Stein J, Murphy DJ, Payne C et al. (2015) A Remedy for fragmented hospital care. Harvard Business Review-New England Journal of Medicine Online Forum: Leading Healthcare Innovation.
2. Stein J, Payne C, Methvin A, et al. (2015) Reorganizing a Hospital Ward as an Accountable Care Unit. J Hosp Med 10: 36–40.
3. Castle B, Shapiro S (2016) Accountable Care Units: A Disruptive Innovation in Acute Care delivery. Nurs Adm Q 40: 14–23.
4. Shapiro S (2015) Accountable care at Emory Healthcare: Nurse-led interprofessional collaborative practice. VOICE of Nursing Leadership 13: 6–9
5. Pronovost P, Berenholtz S, Dorman T, et al. (2003) Improving communication in the ICU using daily goals. J Crit Care 18: 71–75.
6. O’Mahony S, Mazur E, Charney P, et al. (2007) Use of multidisciplinary rounds to simultaneously improve quality outcomes, enhance resident education, and shorten length of stay. J Gen Intern Med 22: 1073–1079.
7. Cowan M, Shapiro M, Hays R, et al. (2006) The effect of a multidisciplinary hospitalist/physician and advanced practice nurse collaboration on hospital costs. J Nurs Adm 36: 79–85.
8. Vazirani S, Hays RD, Shapiro MF, et al. (2005) Effect of a multidisciplinary intervention on communication and collaboration among physicians and nurses. Am J Crit Care 14: 71–77
9. Dowd BE, Greene WH, Norton EC (2014) Computation of Standard Errors. Health Serv Res 49: 731–750.
10. Elixhauser A, Steiner C, Harris DR, Coffey RM (1998) Comorbidity measures for use with administrative data. Med Care 36: 8–27.
11. Volpp KG, Small DS, Romano PS (2013) Teaching hospital five-year mortality trends in the wake of duty hour reforms. J Gen Intern Med 28: 1048–1055.
12. Bilimoria KY, Chung J, Ju MH, et al. (2013) Evaluation of surveillance bias and the validity of the venous thromboembolism quality measure. JAMA 310: 1482–1489.
13. Bosch M, Faber M, Cruijsberg J, et al. (2009) Effectiveness of patient care teams and the role of clinical expertise and coordination: a literature review. Med Care Res Rev 66: 5S-35S.
14. Pannick S, Davis R, Ashrafian H, Byrne BE, Beveridge I, et al (2015) Effects of Interdisciplinary Team Care Interventions on General Medical Wards: A Systematic Review. JAMA Intern Med. 175: 1288–98.
15. Kara A, Johnson C, Nicely A, Neimeier MR, Hui SL (2015) Redesigning inpatient care: Testing the effectiveness of an Accountable Care Team model. Journal of Hospital Medicine 10: 773–779.
16. O’Leary KJ, Wayne DB, Landler MP, et al. (2009) Impact of localizing physicians to hospital units on nurse-physician communication and agreement on the plan of care. J Gen Intern Med 24: 1223–1227.

DOI: 10.31038/JCCP.2019221

Abstract

Background: With the approval of Direct Oral Anticoagulants (DOAC), anticoagulation management has been transformed in both stroke prevention and venous thromboembolism prophylaxis and treatment. Despite evidence-based dosing guidance there has been large variation in prescribing practices that may lead to negative outcomes. The purpose of this two-part study is to evaluate the impact of a DOAC stewardship program on appropriate DOAC prescribing and use.

Methods: Patients were included in part one of the study if they were prescribed initial DOAC therapy from October 15, 2017 – October 15, 2018 with either a general or DOAC specific drug consult. A manual chart review was then conducted for data points including: anticoagulation indication, DOAC dose, serum creatinine, weight, age, consult approval or denial, and reasons for denial. Part two of the study included patients identified through a VA Population-Based Management Tool (PBMT) from January 1, 2018 – September 30, 2018. A manual chart review was then conducted for data points including: flag category, interventions made, and interventions accepted. Patients were excluded in both arms of the study if the duration of DOAC therapy was less than 20 days, incomplete chart review, or if DOAC therapy was prescribed by a non-VA provider.

Results: A total of 592 consults were included in the final analysis in part one of the study. Of the 233 general consults evaluated, 212 (91.0%) were deemed appropriate, 15 (6.4%) inappropriate, and 6 (2.6%) as clinical grey areas. Of the 233 DOAC specific drug consults evaluated, 218 (93.6%) were deemed appropriate, 1 (0.4%) inappropriate, and 14 (6.0%) as clinical grey areas. There was a significant difference in consults worked inappropriately (p=0.0004). A total of 317 PBMT interventions were included in the final analysis in part two of the study. Of those interventions that were actively acknowledged, 233 (95.9%) interventions were completed.

Conclusion: Implementation of a DOAC stewardship program in a healthcare system promotes appropriate and optimal use as well as safety monitoring of DOACs. A drug-specific consult review process improves inappropriate approval or denial of DOAC therapy while the utilization of a population-based management tool efficiently identifies critically important interventions necessary to ensure safe and appropriate use of DOACs.

Introduction

Two disease states that often require anticoagulation for either prophylaxis or treatment are nonvalvular atrial fibrillation (AF) and venous thromboembolism (VTE). Until recently, the primary option for oral anticoagulation was warfarin, a vitamin K antagonist. In 2010, the first Direct Oral Anticoagulant (DOAC), dabigatran, was FDA approved and since then, other DOACs have received FDA approval within the United States (rivaroxaban, apixaban and edoxaban) [1–4]. With the approval of these agents, anticoagulation management has been transformed in both stroke prevention and VTE prophylaxis and treatment. DOACs require no routine lab monitoring to assess anticoagulation effect due to their favorable pharmacokinetic and pharmacodynamic profile when compared with warfarin. However, DOACs are still associated with serious bleeding risks and possess characteristics different from warfarin, including: renal elimination, short duration of action, different drug-drug interactions, administration considerations, and dosing based off of appropriate indication.

Despite evidence-based dosing guidance, there has been large variation in prescribing practices that may lead to negative outcomes. Steinberg and colleagues were the first to analyze the association between DOAC doses and clinical outcomes in patients with nonvalvular AF. Their study, through the ORBIT-AF II Registry, concluded that off-label doses of DOAC therapy for prevention of stroke in nonvalvular AF are associated with increased risk for adverse events [5]. In addition to prescribing practices, adherence to anticoagulants effects patient outcomes. Shore, et al assessed the adherence component of DOAC therapy by specifically studying patients on dabigatran in Veterans Affairs (VA) hospitals. Their study found that one-quarter of patients demonstrated sub-optimal adherence to dabigatran and poor adherence was associated with an increased risk for stroke and all-cause mortality [6]. Dreijer and colleagues determined 8.3% of medication errors from the hospital and primary care settings were caused from anticoagulation agents, with most error reports concerning the prescribing phase of the medication process [7]. With these results and common practices today, it is imperative that prescribers are aware of the need for oversight of DOAC prescribing. Additionally, studies have shown that a pharmacist driven monitoring program improves appropriate prescribing and monitoring of DOACs used for FDA approved indications. Miele, et al. revealed that appropriate prescribing of DOAC therapy was improved after implementing a pharmacist driven monitoring program through a pre- and post-intervention study. They found that 32.4% of doses administered in the pre-intervention group were considered inappropriate, compared to 13.8% in the post-intervention group. Appropriate prescribing included FDA approved indications and appropriate doses of DOAC therapy based on renal function [8].

Tennessee Valley Healthcare System (TVHS) has pioneered a DOAC stewardship program that includes a DOAC drug specific consult to ensure appropriate prescribing at initiation and then long-term surveillance of patients on DOAC therapy by an ongoing population-based management tool. The goal of this stewardship program is to improve appropriate DOAC utilizations by streamlining the initial DOAC consult process along with ongoing management by utilizing anticoagulation clinical pharmacy specialists (ACC CPS) and a DOAC population-based management tool (PBMT). Previously, DOAC therapy was approved through a generic drug consult review process evaluated by a general clinical pharmacy specialist (CPS). This created a lack of consistency between DOAC approval on a patient to patient basis. The newly developed drug consult review process utilizes both a DOAC drug specific consult along with an ACC CPS. TVHS hopes to improve DOAC utilization with regards to appropriate indication, dosing, and safety through a more specific format of DOAC approval. The DOAC PBMT is used within the VA system to allow for continuous review of safety and compliance parameters for optimal care of patients who are prescribed DOAC therapy. The management tool identifies eight flags including: appropriate dosing, valve replacement, notable labs, overdue labs, critical drug interactions, active Non-Steroidal Inflammatory Drug (NSAID) use, overdue refill greater than 4 weeks and renewal due in next 30 days. It is the responsibility of the ACC CPS to actively review the PBMT daily. Upon review, if a flag is present, the ACC CPS will document their assessment in the patient’s electronic medical record; therefore, notifying the provider of a recommendation or change regarding the patient’s current DOAC therapy. In this study, we will evaluate the impact of a DOAC stewardship program on DOAC utilization at TVHS.

Methods

This multi-site, single center, retrospective cohort study was conducted at TVHS, which includes sites in Nashville and Murfreesboro, TN, as well as patients enrolled at any of the healthcare system’s 13 Community Based Outpatient Clinics (CBOCs) throughout Middle Tennessee, southern Kentucky, and northern Georgia.

Patients were included in part one of the study if they had an active DOAC prescription from October 15, 2017 to October 15, 2018 and 18 years of age or greater. Patients were excluded if the prescription was prescribed for short term therapy (inpatient use only and traveling veterans), written by orthopedics, there was an incomplete data set, or if a consult was denied based on criteria for use (CFU). Patients were included in part two of the study if they had an active DOAC prescription from January 1, 2018 to September 30, 2018 with a note title “Anticoagulation Eval and Mgt Secure Messaging” in the electronic medical record indicating an ACC CPS intervention and 18 years of age or greater. Patients were excluded if the prescription was prescribed for short term therapy (inpatient use only and traveling veterans), if the note title was used in error or the patient relocated to a different VA system. The primary endpoint of this study is to determine the appropriate use of DOAC therapy through a drug consult review process. The secondary endpoints are to determine how the drug consult review process and the utilization of a population-based management tool influences safety outcomes of initial and ongoing DOAC therapy including, concomitant antiplatelet therapy in part one and the PBMT flag categories in part two.

Patients who were prescribed direct oral anticoagulation therapy from October 15, 2017 – October 15, 2018 were identified through data warehouse extraction. A manual chart review was conducted for data points including: anticoagulation indication, DOAC dose at time of consult submission, initial or renewal consult, documented labs, documented weight, consult approval or denial, and rational for approval or denial. Evaluated consults were then categorized as appropriate, inappropriate, or clinical grey areas (Appendix 1). Patients followed by the PBMT from March 24, 2018 to September 24, 2018 were identified through data warehouse extraction using the note title “Anticoagulation Eval & Mgt Secure Messaging.” A manual chart review was conducted for data points including the PBMT flag categories (Appendix 2), interventions actively acknowledged and interventions completed. The sample size was calculated based on the primary objective to determine appropriate use of DOAC therapy through a drug consult review process. The sample included all patients who meet study criteria that received DOAC therapy during the pre-determined timeframe. We wanted to see at least a 15% difference regarding the number of appropriate approvals/denials between the generalized and the specialized consult review process. Using these two data points, the effect size was found to be 15% with alpha set at 0.05 and beta set at 0.20. The study required 133 patients in each arm to be adequately powered. Chi square and fisher exact were used to calculate the p-value for the categorical data presented.

Results

Table 1 displays the baseline demographics and clinical characteristics of the two cohorts. The two groups did not different significantly with respect to age, gender, and race. However, there was a statistically significant difference found in use of rivaroxaban and dabigatran between the two cohorts. The authors attribute this difference to the delineated questions of the drug-specific consult. (Table 1)

Table 1. Baseline characteristics for the consult review process

*Indication for Use – “other” includes non-FDA approved indications

Part I: Drug consult review process

Primary Outcomes

Two thousand twenty-one unique patients were extracted from October 15, 2017 – October 15, 2018. Due to time constraints and sample size being met, after the initial data pull the date range was shortened to January 13, 2018 to July 13, 2018. All consults in the generalized cohort were evaluated (n=296) and consults in the specialized cohort were randomized to meet a matching sample size (n=296); therefore, a total of 592 DOAC consults were evaluated. In the generalized cohort, 233 consults were included for study evaluation with 63 excluded for various reasons (incomplete data set, orthopedic patients, denial based on CFU). In the specialized cohort, 233 consults for study evaluation with 63 excluded for various reasons (short term therapy, incomplete data set, orthopedic patient, denial based on CFU). Table 2 illustrates the number of evaluated consults deemed appropriate, inappropriate, and those classified as clinical grey areas. Of the 233 general consults worked, 212 were deemed appropriate, 15 inappropriate, and 6 as clinical grey areas. Of the 233 DOAC Specific Drug consults worked, 218 were deemed appropriate, 1 inappropriate, and 14 as clinical grey areas. Statistical significance was seen only in those consults worked inappropriately. (Table 2)

Table 2. Categorization of consult approval or denial

Of the 15 consults worked inappropriately in the generalized process, inappropriate interventions identified included: non-FDA approved dosing, critical drug-drug interactions, non-FDA approved indications, no indication provided, and hepatic dysfunction. The one consult worked inappropriately in the specialized process, the inappropriate intervention identified was non-FDA approved dosing.

Secondary Outcomes

There were no interventions made in the generalized consult review process regarding concomitant antiplatelet use, however there were 29 recommendations made to either discontinue the P2Y12 inhibitor or discontinue/decrease the dose of aspirin in the DOAC specific consult process. Of the 29 recommendation that were made, 13 (45%) recommendations were completed, specifically 9 (69%) were discontinued and 4 (31%) doses were decreased.

Part II: Population Based Management Tool

One thousand fourteen notes with the title “Anticoagulation Eval and Mgt Secure Messaging” were extracted from March 24, 2018 to September 24, 2018 and randomized using Microsoft excel. Three hundred four notes were evaluated to collect the pertinent information as noted above in the “data collection” section. Two-hundred ninety-two notes were included as 12 notes were excluded for various reasons (orthopedic patient, use of wrong note title, relocation to another VA and no active DOAC prescription). Of the 292 notes that were evaluated, 267 notes only included interventions made on one flag, where as 25 notes included interventions on 2 flags resulting in a total of 317 interventions made.

Table 3 displays the number of interventions made regarding the flag category specified by the population-based management tool. By far, the intervention that was made most commonly was alerting the provider that the prescription needed to be renewed (30%), followed by overdue refill > 4 weeks (22%), overdue labs (15%) and inappropriate dosing (11%). From the 317 interventions that were made 242 (76.7%) interventions were actively acknowledged. Of those that were actively acknowledged, 233 (95.9%) interventions were completed (meaning a medication was renewed, refilled, labs were ordered, medications changed, etc.) with specific percentages regarding each flag noted in table 4. (Table 3, Table 4)

Table 3. Interventions made via the population-based management tool

Table 4. Interventions actively acknowledged and completed

Discussion

As the anticoagulation paradigm begins to shift from warfarin to DOAC therapy, the use of these medications should still be managed and monitored carefully to prevent unwanted harm. To our knowledge, this is the one of the first studies analyzing the implementation of an ACC CPS run DOAC stewardship program that includes a drug-specific consult and a continuous PBMT evaluated by ACC CPS. When using a generalized consult review process, more consults were deemed to be evaluated inappropriately compared to a specialized consult review process. The generalized process included a consult with minimal requirements for completion (drug, dose, and reason why patient was not candidate for preferred formulary agent, dabigatran) and was then reviewed by a general clinical pharmacy specialist. With the implementation of a DOAC stewardship program, a drug-specific consult was created and is now reviewed by an ACC CPS. For completion, the drug-specific consult must include: drug, dose, indication (as well as details associated with indication), appropriate baseline labs (Hgb/Hct, SCr, AST/ALT), bleeding history, antiplatelet use, NSAID use, and use of pillbox.

The results of this study are consistent with previous studies finding that implementing a pharmacist driven monitoring program reduced inappropriate prescribing of DOAC therapy in adult patients with an indication for nonvalvular AF and/or VTE prophylaxis and treatment [8]. In our study, the following inappropriate uses of DOAC therapy were identified: non-FDA approved indications (apical thrombus, cryptogenic stroke), critical drug-drug interaction (strong CYP3A4 and P-glycoprotein inducers), non-FDA approved dosing (lead-in dosing provided for AF, subtherapeutic use of apixaban, use of apixaban with CrCl75 years old or declining renal function, denial of dabigatran due to use of pillbox, and use of rivaroxaban for ease of compliance. By utilizing the PBMT, our study illustrates that continuous monitoring of DOAC therapy is necessary as patient’s clinical status has the potential to change while on DOAC therapy. The authors attribute the one-fourth of interventions not actively acknowledge as due to the lack of education provided to prescribers prior to the implementation of the DOAC stewardship program. However, almost 96% of the interventions that were actively acknowledge were completed, showing the importance of additional surveillance methods rather than relying on prescriber chart reviews. Moving forward, the authors plan to provide educational sessions to prescribers regarding the importance of the PBMT hoping to eliminate alert fatigue and engage providers in the ongoing efforts of the DOAC stewardship program.

There are a few limitations of this study to consider. First, when categorizing the evaluated consults, clinical grey areas have the potential to differ depending on the consult reviewer’s clinical interpretation of current literature. Second, throughout data collection, four reviewers were trained based on the derived protocol; however, due to the abundant number of unique patients and charts to be reviewed data could have been assessed differently between reviewers. Impacts of this study support the need for a DOAC stewardship program in a healthcare system to promote appropriate and optimal use as well as safety monitoring of DOACs. A drug-specific consult review process improves inappropriate approval or denial of DOAC therapy while the utilization of a PBMT efficiently identifies critically important interventions necessary to ensure safe and appropriate use of DOACs.

Appendix 1: Clinical Grey Areas

• Use of apixaban due to age >75 years old
• Denial of dabigatran due to use of pillbox
• LV Thrombus (if failed or cannot take warfarin)
• Weight (>120kg)
• Labs not collected within 90 days
• Use of DOAC versus heparin or enoxaparin for lead-in for treatment of VTE
• Non-FDA approved use of DOACs per Landmark Clinical Trials
• CrCl 15–25 mL/min for use of apixaban in A. Fib
• CrCl 15–30 mL/min for use of rivaroxaban in A. Fib

Appendix 2: Population-Based Management Tool (PBMT) Flags

• Overdue labs
  • Results include all patients where either their most recent hemoglobin OR most recent platelet OR most recent serum creatinine is overdue per patient’s monitoring frequency
  • Monitoring frequency defaults to 12 months for all labs except 6 months for serum creatinine for patients that are 75 years of age or more and/or have a CrCl of less than 60 mL/min
  • Once the overdue lab(s) are resulted (once daily in the morning) the patient will no longer appear in this column. In addition, after review of the patient, monitoring frequency (for serum creatinine) can be reset to 2 weeks, 1, 3, 6 or 12 months as determined clinically appropriate
• Notable labs
  • Most recent platelets <100 x 10⁹ /L if the second most recent value was above 100 x 10⁹ /L OR any platelet value < 50 × 10⁹ /L OR hemoglobin < 10 g/dL or AST/ALT > 135/120 U/L OR a hemoglobin drop > 2 g/dL since previous hemoglobin with resultant value < 13.1 g/dL for males and <11.0 g/dL for females
• Active NSAID use
  • Patients with an active non-aspirin NSAID based on VA Drug Class MS101 and MS102
  • Active includes prescriptions with status ‘ACTIVE’, ‘SUSPENDED’, ‘PROVIDER HOLD’, ‘HOLD’, ‘PENDING’
  • This will include non-VA prescriptions
• Valve replacement
  • Results include all patients with an ICD code for a prosthetic (bioprosthetic or mechanical) heart valve on their problem list, attached to two outpatient visits within the last 2 years, or as inpatient discharge diagnosis within the last two years
• Dosing flag
  • Results include all patients whose renal function (by Cockcroft-Gault with actual body weight) falls above or below specified cutoffs based on agent, indication and if applicable selected drug interactions as per FDA approved package inserts
  • Results also include patients whose dose does not appear to match indication and duration or other non-renal dose modifying characteristics (e.g. high dose apixaban for PE/DVT > 6 months, low dose apixaban for PE/DVT within first 6 months, dose/indication mismatch based on age and body weight)
• Overdue refill > 4 weeks
  • Results include patients for whom it has been more than 4 weeks since their day supply would be expected to run out (based on released date plus one additional week to allow for shipment)
  • Renewal due in next 30 days
  • Patients for whom the ordered days supply is scheduled to expire within the next 30 days
• P2Y12i use
  • Antiplatelet therapy includes aspirin, clopidogrel (Plavix ®), ticagrelor (Brilinta®), prasugrel (Effient ®)
  • Critical drug-drug interactions

• Other critical drug interactions
• Active cancer pharmacotherapy
• Flag if: active VA cancer pharmacotherapy in a patient with diagnosis of DVT/PE ± atrial fibrillation or atrial flutter
• Active includes prescriptions with status ‘ACTIVE’, ‘SUSPENDED’, ‘PROVIDER HOLD’, ‘HOLD’, ‘PENDING’ for any of the interacting drugs or cancer pharmacotherapy
• Non-VA medications will be included in the analysis for critical drug-drug interactions. Patients with non-VA DOAC prescriptions will not be identified by this tool

References

1. Eliquis [package insert]. New York, NY: Pfizer Pharmaceuticals, Inc; 2012.
2. Xarelto [package insert]. Titusville, NJ: Janssen Pharmaceuticals, Inc; 2011.
3. Pradaxa [package insert]. Ridgefield, CT. Boehringer Ingelheim Pharmaceuticals, Inc; 2011.
4. Savaysa [package insert]. Tokyo, Japan. Daiichi Sankyo Company, Limited; 2017.
5. Steinberg B, Shrader P, Thomas L, Ansell J, Fonarow GC, et al. (2016) Off-label dosing of non-vitamin K antagonist oral anticoagulants and adverse outcomes, The ORBIT-AF II Registry. JACC 68: 2597–2604.
6. Shore S, Carey E, Turakhia M, Jackevicius CA, Cunningham F, et al. (2014) Aherence to dabigatran therapy and longitudinal patient outcomes: insights from the Veterans Health Administration. Am Heart J 167: 810–817.
7. Dreijer AR, Diepstraten J, Bukkems VE, Mol PGM, Leebeek FWG, et al. (2019) Anticoagulant medication errors in hospitals and primary care: a cross-sectional study. Int J Qual Health Care 31: 346–352.
8. Miele C, Taylor M, Shah A (2017) Assessment of direct oral anticoagulant prescribing and monitoring pre- and post-implementation of a pharmacy protocol at a community teaching hospital. Hosp Pharm 52: 207–213.