Rema Padman, PhD

 This study illustrates a systematic methodology to embed medical costs into the exact flow of clinical events associated with chronic care delivery. We summarized and visualized the results using clinical and cost data, with the goal of empowering patients and care providers with actionable information as they navigate through a multitude of clinical events and medical expenses.

 We analyzed the electronic health records (EHRs) and medication cost data of 288 patients from 2009 to 2011, whose initial diagnoses included chronic kidney disease stage 3, hypertension, and diabetes.

 We developed chronological pathways of care and costs for each patient from EHR and medication cost data. Using a data-driven method called clinical pathway (CP) learning, which leverages statistical machine-learning algorithms, we categorized patients into clinically similar subgroups based on progressing clinical complexity and associated care needs. The CP-based subgroups were compared against cost-based subgroups stratified by quartiles of total medication costs, and visualized via pathways that are color-coded by costs.

 Our methods identified 3 CP-based, and 4 cost-based, patient subgroups. Two sets of subgroups from each approach indicated some clinical similarity in terms of average statistics, such as number of diagnoses and medication needs. However, the CP-based subgroups displayed significant variation in costs; conversely, large differences in clinical needs were observed among cost-based subgroups.

 This study demonstrates that CPs extracted from EHRs can be enhanced with appropriate cost information to potentially provide detailed visibility into the variability and inconsistencies in current best practices for chronic care delivery.

	This preliminary study extracts patterns and variations in medical spending associated with the flow of clinical events in the management of patients with multiple chronic conditions.

		It illustrates a systematic, generalizable methodology for embedding cost information into the clinical pathways of patient cohorts using electronic health record and cost data.

		It exposes variations in costs among clinically similar patients, and variations in clinical complexity among patients with similar costs, by comparing patient cohorts categorized by their clinical pathways and total medical spending, respectively.

		Future extensions will examine appropriate incorporation of data-driven evidence into shared decision making and innovative cost analyses.

Patient care journeys span the entire continuum of clinical and nonclinical experiences. Navigating through a multitude of these events is a mystifying and daunting experience for most patients, and particularly so for patients with multiple chronic conditions (MCCs).

 Despite being the central figure in their scenarios, research indicates that patients hold little power for steering in an appropriate direction through these passages.

 Although hospitals and individual providers recognize the need to guide patients through their treatments, often such guidance follows a top-down approach and does not actively involve patients in the decision-making process.

 However, with personalized and precision medicine gaining increasing attention, and with more patients and providers participating in the value-based payment models that emphasize patient satisfaction,

 healthcare stakeholders are seeking to facilitate shared decision making by incorporating not only treatment efficacy and clinical factors, but also—equally important&mdash;patients’ personal preferences and financial limitations. In fact, multiple research studies have found that involving patients in the decision-making process can lead to higher satisfaction, lower medical costs, and better health outcomes, among other benefits.

 Prior research also shows that patients’ knowledge of pricing information is associated with lower medical costs for common medical services.

With medical cost being such an opaque subject, providers also may not have the best guidance strategy for the treatments that they offer to their patients.

 Given these challenges, this study proposes a new approach to incorporate medical costs explicitly in the chronologically ordered, clinical pathways (CPs) of patient experiences. We especially focused on MCC patients, who are high utilizers of medical services and often participate in innovative payment programs.

 We illustrated this approach using a cost-centered perspective as well as a clinically focused perspective to show alignment in some subgroups and significant variations in others, in the categorization of pathways and patient subgroups under these 2 differing views. The long-term goal of this research is to eventually achieve accurate predictions of anticipated future events and costs following different clinical and cost pathways for improved shared decision making, and, subsequently, identify appropriate ranges of cost for targeted clinical pathways within a patient population.

CPs, commonly referred to as treatment plans that are time- and task-oriented, describe the essential steps associated with the expected clinical course in diagnosing and treating patients.

 The goals in using CPs are to optimize treatment efficiency, patient outcomes, and medical spending.

 CPs are used widely in the United States and are expected to have an even larger influence on healthcare delivery, quality, and patient outcomes, as bundled or episode-based payments gain momentum. Furthermore, using information from CPs in the operational setting has the potential to optimize not only the efficacy of treatments, but also their costs to patients, providers, and society.

Despite these anticipated benefits, to the best of our knowledge, there is a tremendous gap in the resources that will allow clinicians and patients to incorporate best practices for healthcare delivery in the form of CPs. Therefore, we aimed to begin bridging this gap by developing a data-driven approach that would take advantage of the large amount of detailed, patient-level, clinical information captured in electronic health records (EHRs), coupled with cost data that document the medical costs associated with patients’ clinical activities. This preliminary study applies an advanced, systematic, and generalizable methodology to embed medical costs, specifically, costs of medications, into the exact flow of clinical events associated with the complexities of chronic care delivery. Combined pathways of clinical and cost information are summarized and visualized for patient subgroups following common CPs, and compared against simple cost-based segmentation of the patient population.

CP learning refers to an emerging area of data-driven research to mine the common and rare patterns in the co-progression of 1 or more clinical events, depending on the availability of granular data and types of research questions, using statistical machine-learning methods.

 Our previous studies developed and applied these methods to chronic kidney disease (CKD) to identify CPs containing: a) encounters; b) diagnoses as 

International Classification of Diseases, Ninth Revision

) codes; c) procedures as Current Procedural Terminology (CPT) codes; and d) prescriptions of medications summarized by drug classes.

 This study extends these methods that combine statistical machine learning and data visualization by specifically incorporating costs associated with medications. Developing such detailed visibility into the co-progression of treatments and diseases from actual practice data, rather than abstract clinical guidelines, along with the costs associated with the events—medications, in this instance&mdash;is potentially a useful and novel approach to investigate.

We applied 2 approaches in identifying CPs with cost data. The clinically focused approach used our CP-learning algorithm to detect the differences in progression of clinical factors, but not costs, in order to divide patients into clinically similar subgroups. The costs of medications were subsequently computed for these subgroups. The cost-centered approach, which is analytically simpler, divided patients into similar subgroups based solely on the total amount of estimated co-pays on medications. Generally, the clinically focused approach categorizes patients by pathways of care, whereas the cost-centered approach does so based on the total cost of care.

We collaborated with a community nephrology practice in western Pennsylvania that specializes in the management of CKD. CKD patients—mostly an elderly and vulnerable population&mdash;commonly suffer from a number of serious comorbidities and complications as the disease progresses to end-stage renal disease (ESRD).

 Many CKD patients are Medicare beneficiaries, and the United States Renal Data System reports that CKD patients incur nearly twice the cost for their care, on average, compared with the non-CKD Medicare patient.

 In addition, characteristics of the disease have resulted in limited high-quality randomized clinical trials, further exacerbating the challenges in care delivery.

 Hence, improved management of this population using the approach illustrated in this study may generate multiple benefits, such as reduced costs and increased patient satisfaction.

We obtained detailed clinical data, over the time period from 2009 to 2011, from the EHR extracts of the nephrology practice. We identified 288 patients diagnosed with only CKD stage 3, diabetes, and hypertension, and with no other complications, at the beginning of the 2-year period. The gender ratio of female to male was 0.41 to 0.59, and the average age of patients was 73.4 years (standard deviation [SD] = 10). Race represented by the study patients included Caucasian (94.4%), African American (4.8%), and other (0.8%). Over the 2-year period, patients had, on average, 5.5 (SD = 2.1) office visits, 0.4 (SD = 1.0) hospitalizations, and 0.4 (SD = 0.2) education sessions, respectively.

CPs were created based on all records within 2 years since the initial visit, such that we could track the co-progression of clinical events and associated costs within a fixed time interval. Analysis includes the following patient care information: a) encounter types: office visit, hospitalization, education sessions; b) diagnoses as

 codes: CKD stages 1 to 5, ESRD, hypertension, diabetes, acute kidney injury, hyperparathyroidism, anemia, proteinuria, hyperkalemia, acidosis, hyperphosphatemia, glomerulonephritis, urinary obstruction, volume depletion, rhabdomyolysis; c) procedures as CPT codes: renal Doppler and ultrasound; and d) all medications taken by patients, excluding over-the-counter medications.

In the clinically focused approach, medications are summarized by drug classes for the purpose of patient subgrouping. In the cost-centered approach, total costs are calculated for each medication as is. Therefore, it is possible for 2 patients with similar clinical conditions and needs to be categorized into different levels, such as medium for complexity and high for spending, because CP-based subgroups are based on diagnoses and drug classes, whereas cost-based subgroups are based on costs from co-pays of each medication brand. For the sake of simplicity in illustrating our approach, the analysis in this paper examined only the cost of medications in the form of patient co-pays. We manually obtained the estimated co-pay of each medication from GoodRx that provides information on prescription prices under different insurance plans. To do so, we assumed that all patients were: Medicare beneficiaries, under AARP Medicare Complete Plan 1; residing in zip code 15201 of Pittsburgh, Pennsylvania; being prescribed medication dosages at the lowest appropriate level; and purchasing 30-day medication supplies.

The total estimated medication co-pays over 2 years had a wide range, from $0 to over $18,800, with an average of $1032.70 (SD = $2274.0) under the stated assumptions. In the cost-centered approach, we categorized patients into 4 quartile-based subgroups based on spending: high spenders (75% and up), medium spenders (50%-75%), low spenders (25%-50%), and zero spenders (0%-25%). The clinically focused approach detected 3 subgroups using the CP-learning algorithm based on clinical conditions and needs: high complexity, medium complexity, and low complexity. The 

 displays the descriptive statistics of each subgroup, under both clinically focused and cost-centered approaches.

The Table reveals that the 2 medium subgroups—medium spenders and medium complexity&mdash;obtained from the clinical and cost approaches are in fact clinically quite similar. For example, the subgroups had, on average, 6.2 and 5.4 visits, and 5.6 and 5.3 unique diagnoses, but differed in the number of unique drugs (9.7 vs 3.6, respectively); and as an outcome measure, 16.7% and 14.1%, respectively, of the patients progressed beyond CKD stage 3. Even the average cost, $274.6 and $272.5, respectively, differed by just over $2. However, what is strikingly different is the variation in costs between the 2 subgroups. Whereas the coefficient of variation of medium spenders was 0.73, that of the medium complexity group was 7.1&mdash;nearly a 10-fold increase.

displays the overlap in the assignment of patients to subgroups under clinically focused and cost-centered approaches. Although 97% of the high-spending patients were also part of the high-complexity group, 2 patients in the high-spending group—including the patient who spent the most, $18,801.40&mdash;were in fact assigned to the medium complexity group. One potential explanation, which needs to be verified by the clinicians, is that there might have been excessive spending of medical resources given these patients’ clinical needs, which are at the medium-complexity level.

 (parts a and b) displays the large variations in cost among clinically similar patients. Figure 2(b) displays patients’ actual CPs generated for the medium-complexity subgroup. It can be pictured as a map of visits that patients experienced during the 2-year period. Each node in Figure 2 is a unique visit characterized by encounter type, diagnoses noted, and combinations of drug classes prescribed to patients. Starting with the dense, overlapped areas at the left-hand side of Figure 2(b), with all patients diagnosed with CKD stage 3, diabetes, and hypertension, but no other complications, the fanning pathways show the divergence from the common starting point as patients’ disease progresses and complications emerge in diverse ways, for which varying treatments are provided. The size of the nodes and thickness of the edges reflect frequency of visits and transitions in the data. A larger node suggests that this is a common visit that many patients experience, and a thicker edge is an indication that the transition of visits, and accompanying change in clinical conditions or medication prescriptions, are observed among many patients. The color of each visit represents the spending category in quartiles: high (black), medium (dark blue), low (light blue), and zero (green).

There is growing recognition of the need for more precise risk-adjustment strategies, incorporation of evidence-based treatment variability, and increased use of data and information technology to facilitate patient engagement and shared decision making in promoting value-based payment systems.

 In this preliminary study, we aimed to provide a generalizable framework to estimate the costs associated with actual care delivery, and to expose variations in care and cost. In particular, the results from this study showed significant variation in costs among patients who are clinically similar. A deeper analysis of these pathways may uncover the patterns and causes for these variations within subgroups to allow appropriate incorporation of this evidence into the development of future payment models and care delivery practices.

In this study, we analyzed data on CKD patients whose complex, chronic condition is an example of the high-need, high-cost care delivery context that is a continuing challenge to the healthcare system.

 These patients require coordinated care due to their MCCs, and are a key population to be considered in policy design and implementation.

 Our proposed framework specifically targets this patient population by modeling the co-progression of multiple clinical factors, treatments, and medications. We envision that an information technology—enabled tool based on the demonstrated methodology, once developed, deployed, and rigorously evaluated, can be used at the point of care by clinicians and patients to discuss available courses of treatment options, consider their potential efficacy projected at the cohort and personal level, and, equally important, build awareness of the costs associated with the entire course of treatment. Such tools may also provide policy makers and other stakeholders at healthcare practices access to data-driven evidence for innovative cost analyses.

Although the methodology is generalizable to other health conditions and includes many clinical factors, a major limitation of our analysis is the accuracy and availability of data, particularly relevant to our cost-estimation approach. For illustrative purposes, we assigned all patients to a single Medicare plan and manually obtained the estimated co-pays from a website that provides prescription price information. Therefore, some of the drugs selected and prescribed to patients that were found to be excessively expensive may well be due to the potential discrepancy between our assumed insurance plan and patients’ actual plans.

In addition, variations in CPs observed within each subgroup may be explained by unobserved variables, such as social and behavioral factors, as well as the more extensive health information that was not included in the current analysis. For example, the nature of our EHR dataset limits our ability to infer medication adherence; thus, the generated CPs assume perfect medication adherence, which we recognize to be unrealistic. There is also the possibility that the lack of adherence may lead to divergences in the CPs, but the current data fail to capture such associations. Furthermore, patients’ conditions are gauged using 

codes recorded in the EHR; therefore, we did not distinguish patients by the severity of each condition. For instance, use of insulin, whose choice is a marker of severity in diabetes,

 is often observed among the high-spending/complexity subgroup, but our CP-learning algorithm considered all patients with diabetes to have the same severity. Availability of such detailed, relevant data in the future—such as claims data and lab results&mdash;will help to overcome these limitations.

This preliminary analysis shows that patient subgroups generated by the CP-learning methods may be able to expose variations in costs among patients who are clinically similar, and vice versa, thereby facilitating future research to develop improved treatment plans within innovative payment models.

The authors are grateful to the physicians and staff of the community nephrology practice, who generously provided data from their electronic health record for this study. They particularly thank Pradip Teredesai, MD; Nirav Patel, MD; Qizhi Xie, MD, PhD; and staff members Linda Smith and Audra Barletta, who also gave important clinical and technical information about the data and the key characteristics of chronic kidney disease and its treatments.

 Division of Health Informatics, Department of Healthcare Policy & Research, Weill Cornell Medicine of Cornell University (YZ), New York, NY; The H. John Heinz III College, Carnegie Mellon University (RP), Pittsburgh, PA.

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (YZ, RP); acquisition of data (YZ, RP); analysis and interpretation of data (YZ, RP); drafting of the manuscript (YZ); critical revision of the manuscript for important intellectual content (YZ, RP); statistical analysis (YZ, RP).

 Yiye Zhang, PhD, Assistant Professor, Division of Health Informatics, Department of Healthcare Policy & Research, Weill Cornell Medicine of Cornell University, 425 E 61st St, Ste 301, New York, NY 10065. E-mail: yiz2014@med.cornell.edu.

1. Rein A. Navigating health care: why it’s so hard and what can be done to make it easier for the average consumer. Changes in Health Care Financing and Organization website. http://www.hcfo.org/files/hcfo/HCFONavigatingHealthCare.pdf. Published 2007. Accessed November 1, 2016.

2. Alston C, Paget L, Halvorson G, et al. Communicating with patients on health care evidence. https://nam.edu/wp-content/uploads/2015/06/VSRT-Evidence.pdf. Published September 2012. Accessed November 1, 2016.

3. Jolly SE, Navaneethan SD, Schold JD, et al. Development of a chronic kidney disease patient navigator program. 

2015;16:69. doi: 10.1186/s12882-015-0060-2.

4. Blumenthal D, Abrams MK. Tailoring complex care management for high-need, high-cost patients. 

. 2016;316(16):1657-1658. doi: 10.1001/jama.2016.12388.

Articles

The authors illustrate a methodology for delineating variations in medical costs for patients with similar clinical conditions and needs using electronic health record data.

Data-Driven Clinical and Cost Pathways for Chronic Care Delivery

Chronic diseases are common, complex, and expensive health conditions that can benefit from innovations in healthcare service delivery enabled by information technology and advanced analytic methods. This paper proposes a data-driven approach, illustrated in the context of chronic kidney disease (CKD), to develop clinical pathways of care delivery from electronic health record (EHR) data. 

We analyzed structured and de-identified EHR data from 2009 to 2013 of 664 CKD patients with multiple chronic conditions.

Machine learning algorithms were used to learn data-driven and practice-based clinical pathways that cluster patients into subgroups and model the co-progression of their encounter types, diagnoses, medications, and biochemical measurements. Given a pattern of biochemical measurements, our algorithm identifies the most probable clinical pathways, and makes predictions regarding future states, with and without temporal information. CKD stages, their complications, and common medications are included in the clinical pathways.

Using the EHR data of 664 patients who were initially in CKD stage 3 and hypertensive, we identified 7 patient subgroups—each distinguished primarily by the type of complications suffered by the patients. Our algorithm demonstrates fair accuracy (up to 44% and 75%, respectively) in learning the most probable clinical pathways and predicting future states associated with temporal patterns of biochemical measurements and patient subgroups. 

Data-driven clinical pathway learning summarizes multidimensional and longitudinal information from EHRs into clusters of common sequences of patient visits that may assist in the efficient review of current practices and identifying potential innovations in the care delivery process.

		The availability of high-volume, time-stamped, and individual-level health data is beginning to facilitate clinical interventions that are personalized and predictive. 

		Healthcare service delivery can benefit greatly from advanced statistical and machine learning models and algorithms that can learn personalized insights from electronic health record (EHR) data. 

		Data-driven clinical pathways that describe the co-progression of encounter types, diagnoses, medications, and individual biochemical measurements can be learned from EHR data, using statistical and machine learning methods to support the review of current practices and innovate healthcare delivery approaches. 

		Our proposed methodology is generalizable to other clinical conditions and can accommodate varying numbers of clinical and other relevant factors.

  According to the World Health Organization, 60% of all deaths, worldwide, can be attributed to chronic diseases such as diabetes, heart disease, stroke, and cancer; they are also a major cause of poverty and lack of economic development.

 As part of a multi-pronged effort to address this challenge, innovations in chronic care delivery are beginning to leverage advanced statistical and machine learning models and algorithms to obtain new insights into care quality, outcomes, and cost.

 Machine learning is the science of constructing algorithms that learn from large volumes of data in order to facilitate decision making by generating potentially new insights; it has gained widespread implementation across many industries today.

 Just a few examples of machine learning applications are speech recognition, self-driving cars, and personalized online experiences.

Although innovations driven by machine learning have seen tremendous success,

 subsequently resulting in improved service performance, productivity, and growth,

 for a variety of reasons, the healthcare industry has been relatively slow to incorporate these techniques into decision-support applications and to adapt to resulting changes.

 For instance, in making treatment decisions, many clinicians may prefer to use clinical practice guidelines (CPGs) over predictions generated by machine learning algorithms—algorithms which may seem like a “black box” with little relevance to actual clinical decision making.

 However, many of the current clinical decision support capabilities, whether CPG-embedded electronic health record (EHR) interactivity or computerized provider order entry (CPOE) application, are designed by humans and target the “average patient.”

As the Precision Medicine initiative states,

 we are now in an era in which clinical interventions need to be personalized and predictive, and so should decision support recommendations. To meet this objective, it is no longer sufficient to rely on CPGs, often created based on consensus opinions or randomized clinical trials that have strict enrollment criteria. Rather, with the tremendous amount of data being accumulated in EHRs from the enactment of the Health Information Technology for Economic and Clinical Health (HITECH) Act as part of the American Recovery and Reinvestment Act,

 healthcare service delivery can also benefit greatly from advanced statistical and machine learning models and algorithms that can learn potentially useful insights from large amounts of highly detailed data collected daily, as part of routine care delivered in multiple, diverse settings.

Traditional topics in machine learning include classification and unsupervised learning.

 Classification refers to the method of labeling unknown data to target variables through training a classification model using labeled data. Logistic regression and naïve Bayes are examples of classification algorithms.

 For example, Lee et al used logistic regression to predict 7-day mortality from heart failure in emergent care using initial vital signs, clinical and presentation features, and laboratory tests.

 Unsupervised learning refers to the identification of latent groups in the data. Unlike classification, which is also called “supervised learning,” unsupervised learning does not have true labels, typically does not have true labels, and users need to predefine the number of latent groups. K-means and hierarchical clustering are 2 of the most common unsupervised learning algorithms.

Zhang et al used a variant of the K-means clustering algorithm to design more efficient order sets from historical order data in a pediatric inpatient setting.

 Order sets are groups of relevant orders traditionally clustered together by clinical experts and used within CPOE; this is an example of a manually designed healthcare information technology application that requires significant labor- and knowledge-intensive effort for maintenance and update. In the same study, Zhang et al demonstrated that order sets can be created using machine learning algorithms, with the resulting data-driven order sets requiring less physical and cognitive workload in usage because the methods were trained to find the optimal combinations of orders that matched, with order data generated from actual work flow. In addition to these classical approaches, many advanced machine learning algorithms have been developed and applied over the years to facilitate a more efficient, safer healthcare system.

In this paper, we present a machine learning approach for learning the most probable, data-driven clinical pathways from the EHR data of patients with chronic kidney disease (CKD), and predicting the most probable upcoming interventions at any stage, given recent history. CKD is a chronic condition that currently affects more than 26 million US adults, with an additional 73 million at increased risk for the disease.

 It is also associated with increased risk for cardiovascular disease and acute kidney injury (AKI), and the majority of the patients also suffer from comorbidities such as hypertension and diabetes.

 Consequently, CKD management is complex and expensive, and a large proportion of the US Medicare budget every year is allocated for the treatment of CKD.

 Specifically, the per person per year average cost of treating CKD was $23,128 in 2011—more than twice the average cost of treating non-CKD conditions in the Medicare population ($11,103).

 With the cost increasing and quality of life decreasing as the disease progresses to end-stage renal disease (ESRD),

 there is a growing imperative to pursue innovations in service delivery and management of CKD and other chronic conditions that may generate improved health outcomes, cost savings, and patient satisfaction.

Additionally, generating the highest quality scientific evidence and associated practice recommendations for chronic conditions such as CKD is a continuing challenge for the healthcare field.

 One of the most recent CPGs for CKD was published by the National Kidney Foundation’s Kidney Disease Outcomes Quality Initiative in 2012, which is an update of its 2007 guideline. However, of its 7 key recommendations, only 2 recommendations received the highest grade from the Evidence Review Team of the guideline Work Group for strength of recommendation (“recommend” vs “suggest”), and the highest grade for quality of evidence (“high” vs “moderate,” “low,”  “insufficient”), while other recommendations received lower grades for strength of recommendations and for the quality of evidence.

In this paper, we propose that evidence from actual practices, particularly those that include large number of patients in local treatment settings over reasonable durations, may be used to assist guideline development. We present methods for knowledge extraction from data using machine learning algorithms, and demonstrate that such knowledge can be regarded as practice-based, data-driven clinical pathways. Clinical pathways translate CPG recommendations into an actionable plan such as flow charts, and are used by more than 80% of US hospitals for at least 1 intervention.

 This research aims to develop clinical pathways not strictly based on CPGs, but practice-based evidence learned from data. An overall framework of our approach that supports a learning healthcare system is presented in 

Data-driven clinical pathway learning has garnered research interest since the 1990s,

 but there is limited research on machine learning approaches for the problem. Recently, Lakshmanan et al used a type of clustering algorithm, called DBScan, to cluster patients’ history prior to pathway learning, and applied SPAM, an algorithm to find frequent patterns in pathways, to associate patterns with patient outcomes.

 Huang et al used topic model, a recently developed probabilistic method, for learning latent topics from documents, to discover clinical pathway patterns from EHR event logs.

 Zhang et al modeled clinical pathways as Markov chains that included the co-progression of multiple interventions and diagnoses, and visualized them to allow identification of variations in care and outcomes across latent patient subgroups.

In this paper, we combine clustering and temporal modeling to elicit common clinical pathways from the data. Specifically, given patient characteristics and a sequence of laboratory observations from multiple laboratory tests, we illustrate methods to learn the most probable sequence of clinical interventions that are associated with the laboratory observations, and to make predictions about patients’ impending conditions as a result of the interventions. This approach allows us to link patients’ biochemical responses with clinical interventions and with specific outcomes, thus providing a novel methodology for data-driven clinical pathway learning.

To accommodate the heterogeneity in the patient population and improve model accuracy, we group patients according to similarity of their clinical history prior to pathway learning and prediction. We expect patients’ pathways to branch out as their health conditions and corresponding treatments evolve in different ways. Therefore, prior to pathway learning and prediction, we use hierarchical clustering to cluster patients’ pathways into subgroups according to longest common subsequence (LCS) distance measure.

 LCS is the longest subsequence that 2 sequences have in common, while preserving the order of occurrence of the items in the sequences, but items are possibly separated. LCS has been widely applied in biomedical research as a similarity measure used in trajectory analysis and protein sequence analysis.

 The distance measure, dLCS, is then computed as the difference between the sum of the lengths of 2 sequences and twice their LCS. (Details are in the 

.) Hence, dLCS is affected by the length of the identified subsequence, and the lengths of both sequences; for example, given the same length of LCS, dLCS is bigger for 2 long sequences than 2 short sequences. Therefore, clustering using dLCS allows us to group patients who not only share similarity in clinical interventions, but also have similar durations of treatment. The optimal number of clusters is determined using Silhouette, a measure commonly used in cluster analysis.

 In this study, we consider clusters that have 10 or fewer patients as outliers, and plan to evaluate rare events and exceptions in future research.

 illustrates our modeling scheme for learning the clinical pathways. Given the time stamps associated with intervention data recorded in the EHR

 we assume that each state in the data-driven clinical pathway is separated by at least 1 time unit (eg, day, week, month), and that each state may contain more than 1 type of intervention. For example, it is typical for a CKD patient to have a follow-up visit in the clinician’s office, receive medication prescriptions, and have diagnostic codes assigned to the visit. Our data encoding anticipates such multidimensional and longitudinal features in the data. We assign a unique label for each unique combination of interventions occurring from a visit on the same day, such that patients’ clinical interventions that span multiple categories, such as diagnosis, medication prescription, and encounter type, can be transformed into 1-dimensional pathways, as shown in the top row in Figure 2. Naturally, these interventions are related to one another over time in varying degrees. For instance, interventions that occurred within 6 months of each other may be more strongly correlated than those that occurred within 2 years of each other.

In the context of CKD management, we assume that interventions at visit t+2 are dependent on activities at visit t+1 and t, as shown in the middle row in Figure 2. For analytical tractability, and reflecting actual practice in the management of many health conditions, the time intervals between 2 consecutive visits are categorized as: 1) less than 3 months, 2) greater than 3 but fewer than 6 months, or 3) at least 6 months. These assumptions are practice- and condition-specific,

 but can be readily modified for different settings. Patients’ biochemical conditions, as reflected by their laboratory observations, are assumed to be influenced by the interventions, as shown in the bottom row in Figure 2. For the problem of clinical pathway learning described in this study, our goal is to learn the most probable sequence of clinical interventions given to patients with a particular trajectory of biochemical responses. Similarly, the prediction problem is to infer the most probable imminent interventions in the next state—most importantly, diagnostic codes&mdash;for these patients.

We model this treatment process as a hidden Markov model (HMM). HMM is a statistical model with a wide range of applications, such as in speech recognition and RNA sequence analysis.

 It is defined by 5 elements: sequence of hidden states, sequence of observations, state transition probability distribution, observation probability distribution, and initial state distribution.

 HMM is used to represent a process in which a sequence of observations is generated, and each observation is triggered by an underlying process that is hidden to us. For example, given a sequence of a patient’s body temperatures, we may assume that the patient’s health condition is affecting his or her body temperature. Therefore, the sequence of body temperatures form the observations in HMM, and health conditions represent HMM’s sequence of hidden states.

The sequence of hidden states in an HMM has a first-order Markov property, which states that the current state only depends on the previous state.

 Therefore, we regard the middle row in Figure 2 as the sequence of hidden states and the bottom row as the sequence of observations. Parameters of the HMM, such as transition probabilities of hidden states in the Markov chain, are learned from the data using the expectation-maximization (EM) algorithm.

 Given HMM parameters, we can perform both the clinical pathway learning and prediction tasks through HMM decoding, which calculates the sequence of hidden states with the highest probability given the sequence of observations and the probability distribution of the model. Details of the model and algorithm are described further in the eAppendix and prior studies.

We demonstrate the methodology using a real-world data set of 664 patients, with visits from 2009 to 2013 extracted from the EHR, who suffered from CKD and associated complications. The gender ratio is nearly equal. Over 67% of the patients are aged at least 70 years, and nearly 95% are Caucasian. Components considered as part of clinical pathways and the number of unique patients who had each component in their EHR are listed in 

. These components were selected for their relevance in CKD management, per consultation with clinicians, but can be extended to include additional details. All 664 patients had initial diagnoses of CKD stage 3 and hypertension, but not diabetes, and none of the patients had anemia or hyperparathyroidism initially. These patients either progressed to advanced CKD stages and ESRD, or improved to CKD stages 1 and 2. Most of them subsequently developed some of the complications listed in Table 1.

, was determined to be 7 using the highest silhouette value (0.189) from hierarchical clustering. 

 describes the characteristics of each group in detail, indicating that hierarchical clustering using dLCS was able to divide patients into subgroups that differ on treatment frequency, duration, and outcome at the end of the study period. For example, 95% of the patients in subgroup 5 showed improvement in their conditions at the end of the study period, while none worsened, after being in the clinic for an average of 26.9 months. Subgroup 3 is the largest subgroup, and it also has the smallest average dLCS, suggesting that patients are more similar to one another compared with other subgroups. Subgroup 2, which needs to be investigated further, had a mixture, with 14% of patients who improved and 20% who worsened. The final column in Table 2 lists complications of CKD that the majority of patients suffer from in each group.

 summarizes the accuracies associated with predicting the imminent interventions and diagnoses, such as prescription of diuretics and episodes of AKI, and learning the most probable pathways for sample subgroups 3, 4, and 5. We chose these 3 subgroups because of their larger subgroup sizes, and interesting final outcomes at the end of the study. We tested the accuracies using the most common sequence of laboratory observations (LOs) from 3 consecutive visits, and the number of patients who experienced such patterns is listed under the column, “Number of patients who had LOs.” Training and testing were performed through a variant of the leave-one-out cross-validation method.

Learning and prediction were done with respect to the most common sequence of LO in each subgroup. It is interesting to note that the common biochemical patterns in subgroups 3 and 4 are the same, but the model identified different clinical pathways for these 2 groups, which require further examination. “Pathway with time”/“Pathway without time” measure accuracy of learning the entire pathway, including/not including the actual time duration between 2 visits, respectively. Similarly, “Future visit with time”/“Future visit without time” measure the prediction accuracy for patient’s future interventions, with/without time durations between visits. Each state variable contains information on the presence or absence of 3 encounter types, 19 diagnoses, and 4 drug classes, in addition to 3 different durations between visits. Therefore, the probability of accurate learning and prediction, on a random try, is extremely low compared with the results from our algorithm.

We also examined the false negative and false positive rates in the prediction of an imminent condition such as AKI. We define a false negative to be a case where patients’ CKD stages are worse than predicted, or patients developed AKI, which our methods failed to predict. A false positive is defined as patients’ CKD stages being better than predicted, or prediction of AKI when no AKI developed in reality. We include AKI in this analysis because it is a serious adverse outcome: it often requires hospitalization and can be fatal.

 We were able to obtain false-positive and false-negative rates that are as low as 0%, although this result needs to be validated using a much larger sample. Nevertheless, the learning and prediction algorithms show promise in identifying common pathways of treatments, but these need to be analyzed further to better delineate effective interventions in the various subgroups.

This paper provides a brief overview of machine learning approaches to assist medical decision making, and introduces a methodology, as well as an application that illustrates the development of data-driven clinical pathways through mining of EHR data. This approach may facilitate timely extraction of potential new evidence that could become the basis for new clinical trials, and may also serve as “shared baselines” to be used within a local practice for work flow and population health management.

 Patient-focused applications derived from our research, particularly those that visualize the clinical pathway and provide related patient-oriented recommendations and educational resources, may enhance patients’ understanding of their diseases and treatments, thus facilitating shared decision making.

An important ongoing study is to develop prediction models for other significant outcomes of interest in the management of CKD and its complications. Also, we need to evaluate these data-driven clinical pathways, especially their divergence and rare events, and their predictions with input from clinical professionals. As a growing number of healthcare organizations pilot new care delivery and payment models, such as the accountable care organizations,

  exploring disease trajectories that incorporate the interactions of clinical interventions and their associated outcomes may also provide useful insights on the cost effectiveness of treatments, which organizations can leverage for implementing innovative care delivery practices.

A crucial prerequisite for success in the application of advanced machine learning methods to healthcare delivery is data quality. It is not uncommon for computational scientists to spend significant effort in cleaning EHR data before analysis. In addition, even after months of processing, there are often still missing data and errors, some arising from the mismatch between actual work flows and process assumptions, subjecting the analytical results to bias. Such inefficiency can be minimized by careful observation and understanding of the care delivery context, and planning of the data storage with a range of options available depending on the data size.

 At the same time, methods have been developed, such as imputation and approximate inference algorithms, that can accommodate missing data. For example, in this paper, we used the EM algorithm to infer the parameters of HMM. Furthermore, diversity is innate to most healthcare data, and we found it to be one of the biggest challenges in accurately inferring clinical pathways, requiring large amounts of data and robust methods for analysis and inference. In this paper, we examined encounter type, diagnosis, medication prescriptions, and biochemical measurements, but our data representation is flexible with regard to the number of clinical factors of interest. Therefore, when sufficient curated data becomes available, factors such as medical expenses and behavioral information can also be incorporated to enrich the learned pathways and personalized predictions of health and cost outcomes.

This paper presents additional promising evidence of the potential of machine learning applications for clinical decision making. We develop and demonstrate a methodology to facilitate more targeted management of patients with complex chronic conditions using data-driven clinical pathways. Clinical pathways are learned from a healthcare organization’s EHR data by summarizing multidimensional clinical history as chronologically organized sequences, capturing information on the co-progression of encounter types, diagnoses, medications, and biochemical measurements. Further, we link clinical pathways to a few outcomes within subgroups of patients with reasonable accuracy using hierarchical clustering and HMM. Applying our methodology to relevant EHR data on 664 patients with CKD stage 3 and hypertension, we identify clinical pathways that may be compared with current CPG recommendations in future studies, and contribute to the development of shared-baseline within hospitals. These methods and broad findings from EHR data are generalizable and can be adapted to other clinical conditions to support efficient review of treatments and outcomes and to aid clinical professionals and patients in making more informed treatment and management decisions.

This paper demonstrates that data-driven clinical pathways can be developed using electronic health record data to facilitate innovations in practice-based care delivery for chronic disease management.

Rema Padman, PhD

Articles

Data-Driven Clinical and Cost Pathways for Chronic Care Delivery

Innovations in Chronic Care Delivery Using Data-Driven Clinical Pathways