Abstract
Background
Interventions to support patients with complex needs have proliferated in recent years, but the question of how to identify patients with complex needs has received relatively little attention. There are innumerable ways to structure inclusion and exclusion criteria for complex care interventions, and little is known about the implications of choices made in designing patient selection criteria.
Objective
To provide insights into the design of patient selection criteria for interventions, by implementing criteria sets within a large health plan member population and comparing the characteristics of the resulting complex patient cohorts.
Design
Retrospective observational descriptive study.
Subjects
Patients identified as having complex needs, within the membership population of Kaiser Permanente Southern California, a large, population-based health plan with more than 4 million members. We characterize five commonly used archetypes of complex needs: high-cost patients, patients with multiple chronic conditions, frail elders, emergency department high-utilizers, and inpatient high-utilizers.
Measures
We selected an initial set of criteria for identifying patients in each of the archetypical complex populations, based on available administrative data. We then tested multiple variants of each definition. We compared the resulting patient cohorts using univariate and bivariate descriptive statistics.
Key Results
Overall, 32.7% of the 3,112,797 adults in our population-based sample were selected by at least one of the 25 definitions of complexity we tested. Across definitions the total number of patients identified as complex ranged from 622,560 to 1583 and complex patient cohorts varied widely in demographic characteristics, chronic conditions, healthcare utilization, spending, and survival.
Conclusions
Choice of patient population is critical to the design of complex care programs. Exploratory analyses of population criteria can provide useful information for program planning in the setting of limited resources for interventions. Data such as these should be generated as a key step in program design.
Similar content being viewed by others
INTRODUCTION
Patients with complex needs are heterogeneous in their characteristics and vary in their needs.1 The question of how to identify patients with complex needs and match them with appropriate interventions is a persistent subject of debate within the emerging complex care field.2,3 While evidence about the design of effective interventions for populations with complex needs is slowly accumulating,3,4,5 the research literature has paid less attention to the implications of patient selection decisions for complex care interventions. When the patients selected for an intervention are a poor fit, the result can be low patient uptake of the intervention, inability of intervention services to meet patient needs, and failure to achieve program objectives such as cost or utilization goals.
High-level taxonomies describe the major groups of complex patients. Most prominent is the taxonomy put forward by the National Academy of Medicine (NAM),3 which categorizes complex patients according primarily to their health and clinical conditions, with social risk factors and behavioral health needs that are cross cutting. In practice, complex care programs often use readily available data to select patients for interventions, by specifying a set of inclusion and exclusion criteria. While such criteria sets are sometimes intricate (such as the “impactibility score” described by DuBard et al. in 2018) 6, many patient selection methods in the complex care literature involve one or two criteria defined using simple thresholds.7 A recent review surveyed the complex care literature to characterize the breadth of current approaches to defining complexity. Among the 90 populations with complex needs reviewed, the authors found no two population definitions that were identical.7 While the authors identified common themes in the approaches used to define complex populations, they also noted that there are innumerable ways in which these definitions can be specified. Even within one conceptual approach (e.g., high-cost patients), population criteria may differ in timeframe, thresholds, and other specifications.7 It remains unclear whether seemingly nuanced differences in criteria specifications have a meaningful impact on the characteristics of the resulting populations.
In this study, we used population-based data from a large integrated healthcare system to compare multiple approaches to selecting patients with complex needs. We selected five archetypical complex care patient definitions that are common in the literature, either appearing in the NAM complex population taxonomy 3 or frequently represented in the survey of complex patient definitions by Davis et al.7 We then implemented inclusion and exclusion criteria in an adult population and explored the characteristics of the patient cohorts. We aimed to provide insight into the implications of definitional choices by comparing cohorts of complex patients both within archetypes and between archetypes. We argue that descriptive exploration of complex population definitions provides valuable insights for design and implementation of effective interventions for patients with complex needs.
METHODS
Study Design, Setting, and Population
This retrospective observational study included adult members enrolled in Kaiser Permanente Southern California (KPSC), an integrated health plan and healthcare system that serves more than 4.5 million members. The enrolled population is broadly representative of the ten counties that comprise southern California.8 The institutional review board of KPSC approved this data-only study.
Data Sources
Data were extracted from existing health plan and care delivery sources including membership records, integrated claims, ambulatory and hospital-based electronic health records (EHR), mortality records, and other administrative data.
We assembled patient data from calendar years 2015 and 2016. We used calendar year 2015 as the “baseline” period in which we applied study definitions to identify patients with complex needs. Calendar year 2016 was then used to describe the characteristics of these patients with complex needs in the subsequent year.
Study Definitions
We included as eligible patients those who were enrolled as KPSC members during December 2015, were alive, and were aged 18 or older as of December 31, 2015. We did not require continuous enrollment in either study year.
We selected the following five archetypical conceptual approaches to identifying patients with complex needs: high-cost patients, patients with multiple chronic conditions (MCCs), frail elders, high utilizers of the emergency department (ED), and inpatient (IP) high utilizers.
For each conceptual approach to identifying patients with complex needs, we implemented an existing set of inclusion criteria based on methods found in the complex care literature as follows: We defined high-cost adults by their total annual spending as a percentile of the adult population, using the commonly applied top 5% threshold.9 The MCC concept was adapted from the definition established by Joynt and colleagues and cited by NAM,3 using the presence of 3 or more complex chronic conditions or 6 or more non-complex chronic conditions, from a list of 29 conditions.10 To identify frail elders, we used age of 65 or greater combined with 2 or more frailty indicators, also based on Joynt and cited by NAM.3,10 We identified ED high utilizers based on total number of ED treat and release encounters per year; the threshold was selected for our population based on application of the method developed by Locker and colleagues.11 Finally, we identified IP high utilizers based on total number of hospital admissions per year (including observation stays); we selected a threshold of 2 or more inpatient admissions in the prior year, based on the definition used by Newman and colleagues.12
After establishing the core definition for each conceptual approach to identifying patients with complex needs as described above, we created four additional variants of each definition by adjusting the thresholds used to include/exclude patients, for a total of 5 variants of each concept. This method resulted in a total of 25 cohorts across 5 definitional concepts. The cohorts were not mutually exclusive, either within a concept or between concepts. Table 1 lists the five concepts of complexity, the 25 variants, a shorthand name for each variant, and a brief definition of the concept. Detailed criteria specifications for each definition are provided in the supplemental materials, including International Classification of Diseases 9 and 10 (ICD-9 and ICD-10) codes and Durable Medical Equipment (DME) codes for all chronic condition and frailty indicators (Online Supplement).
Statistical Analyses
We identified all patients who met each definition based on their utilization and characteristics in 2015. We described patients in each of the 25 cohorts by providing information about basic demographic characteristics, survival through the end of 2016, and number of chronic conditions diagnosed during 2016 (among the 29 conditions defined previously). We then characterized patients’ annual rate of IP, ED, and ambulatory (office) visits in 2015 and 2016 expressed as a rate per thousand members per month (PTPM) to account for length of membership in each time period. Finally, we described patients according to their spending trend across the two years, using total annual spending in 2015 and 2016. We classified all patients into four mutually exclusive spending trend groups: persistent top 10%, never top 10%, cost “bloomers” with a rising cost trajectory,13 and cost “wilters” with a falling cost trajectory. To establish these groups, we calculated the 90th percentile for total spending in each year. We then classified each patient’s spending in each year as above or below that threshold. Patients were in the “persistent top 10%” group if their spending was in the top 10% in both years, while those whose spending was below the 90th percentile threshold in both years were in the “never top 10%” group. “Cost bloomers” progressed from lower deciles in 2015 to the top 10% in 2016. Conversely, we defined “cost wilters” as patients who were in the top 10% in 2015 but regressed to lower deciles in 2016. As a sensitivity analysis, we tested the result of establishing the cost-trend groups based on costs standardized for enrollment time, calculated as per member per month spending. The results were not materially different and are provided in the supplemental materials (Online Supplement).
We tabulated or graphed the descriptive measures to visualize the differences within and between complex patient concepts. Because of the very large sample sizes in our study, and because our cohort definitions were overlapping both within and between concepts, we did not use tests of statistical significance. We described continuous variables with means and standard errors, and categorical variables with frequency and percentages. All analyses were performed using SAS 9.4 and visualized in Tableau.
RESULTS
Of the total 3,112,797 adult members of KPSC who were included in the overall study population, 32.7% were selected by at least one of the definitions of complexity we tested based on their characteristics in 2015 (Table 2). The remaining 67.3% were not classified as having complex needs according to any of the population definition variants we examined.
The high-cost patient concept resulted in the largest cohorts of complex patients with 622,560 patients in the “Top 20%” variant (Table 2). The MCC, frail elders, and IP high utilizer groups had the greatest proportion of older adults. Gender distribution was relatively consistent between the patient concepts, but within all concepts except the ED high utilizer concept, the percent female decreased as the definitional variant became more stringent. The percent White Non-Hispanic generally increased within each concept as the definition became more stringent. Survival throughout the follow-up year dropped off as the definitional variant became more stringent. One-year survival was worst among frail elders, with 90% surviving in the frailty 1+ group compared to only 70% surviving in the frailty 5+ group (Table 2).
The distribution of the number of chronic conditions varied both within and between concepts (Fig. 1). The burden of comorbidity was highest in the MCC variants, but frail elders had a strikingly similar level of comorbidity despite being selected based on criteria focused on frailty rather than chronic disease. The least intensive burden of comorbidity was found in the ED high utilizer concepts. Even in the group with 5 or more ED visits during 2015, a third (32%) had 3 or fewer chronic conditions (Fig. 1).
Utilization of IP, ED, and ambulatory services in 2015 and 2016 is shown in Figure 2, which displays PTPM rates of utilization for the two years as a slope graph, where a downward slope denotes an average decrease in rate of utilization over the two years, and an upward slope denotes an increase. In the top 1% high-cost patient cohort, IP utilization into the follow-up year declined very slightly, but the rate of ED visits increased from 160 PTPM in 2015 to 204 PTPM in 2016; the rate of ambulatory visits decreased over the same time period. Similar trends were observed for those with the most chronic conditions (MCC 6) and in the most frail group (frailty 5+). As concept variants become less stringent, rates of utilization were lower but the same general trends in IP, ED, and ambulatory visits were found. By definition, the ED high utilizer groups had more extreme rates of ED utilization; some regression to the mean in rate of ED utilization was seen in all but the most extreme group (ED 5+), and IP utilization was trending upward for all of the ED high utilizer variants. Similarly, the IP high utilizer variants had substantially higher IP utilization rates than any of the other concepts; within the IP high utilizer concept, rate of IP utilization declined in the second year, while ED utilization increased over the same period.
Figure 3 shows the proportion of patients in each concept variant cohort having persistently high costs (dark orange), rising costs (“bloomers,” light orange), falling costs (“wilters,” light gray), or persistently lower costs (dark gray) over the two-year period. Across all of the concepts and particularly in the least stringent variants, a substantial proportion of patients were in the lower 90% of spenders in the second year (never top 10% or cost wilters). Within the top 10% cohort, 57% had “wilting” costs in the second year and 43% persisted in their top 10% level of spending (Fig. 3). Across concepts, the breakdown of spending trend was relatively similar, although the ED high utilizer variants tended to have fewer persistent high-cost or cost “blooming” patients than in the other concepts.
DISCUSSION
There is considerable variation in methods to identify patients with complex needs 7 and common terms found in the literature (e.g., high-cost patients or high utilizers) are somewhat ambiguous. In this study, we implemented simple definitions for five common conceptualizations of complexity, and then explored the effect of a range of criteria specifications within each concept to characterize population differences that are apparent between concepts. We observed that choice of population definition criteria can have a substantial impact on population size and characteristics. Across all 25 definitions of complexity in our analyses, the total number of patients identified as complex ranged from 622,560 (top 20%) to 1583 (frailty 5+). This alone has implications for intervention design, as programs serving very large patient populations may be infeasible unless they are technology-based or systems-level interventions. Within each concept, the choice of more stringent thresholds for population selection led to progressively smaller populations.
Using descriptive methods, we uncovered many other differences between the complex populations we assembled, including differences in mortality, comorbidity, utilization level and trend, and spending level and trend. Within each concept, patterns of increasing morbidity, mortality, utilization, and costs were consistently seen as the population criteria became more stringent. This is intuitive and expected. More notable are patterns found in the comparisons between concepts.
Planners of complex care interventions can benefit from considering the hypothesized impact of their intervention in the context of the target population’s baseline conditions and utilization. For example, the cohorts we assembled based on high ED utilization had the lowest level of comorbidity among all of the complex patient concepts we tested, but their level and trend of utilization was relatively comparable to the cohorts with multiple chronic conditions (MCCs). This observation might lead program designers to explore the patient needs that are driving utilization in these groups and tailor an intervention to amplify those services that are responsive to population needs and characteristics. Alternatively, program designers might adjust population selection approach to more precisely target patients who are likely to benefit from planned services, maximizing the fit of the study population to a planned intervention, such as narrowing the focus of an ED high utilizer intervention on those with frequent ED visits and chronic conditions.
Considering trends in utilization and costs while planning an intervention is also important because of the potential risk of mistaking a pre-post change in population outcomes for a program effect (or failure). For example, a program intending to engage patients with high costs with the goal of reducing total spending might mistakenly interpret patterns of wilting costs as evidence of a program effect in a simple pre-post analysis. Review of descriptive data and recognition of the existing pattern of wilting costs at baseline could cause program planners to avoid this issue by leveraging a more robust evaluation design, such as a difference-in-differences analysis. Another approach could be development of predictive criteria to identify currently high-cost patients who are likely to have wilting costs in the future (and exclude them from the intervention population). While this analysis did not explore the predictability of future spending and utilization patterns in our population, others have demonstrated that such prediction is possible 13,14 and have reflected on the potential advantages of such an approach 15.
Our study had limitations. We selected a handful of concepts for patients with complex needs, and implemented variations of each one. Our selected population definitions are not a comprehensive survey of all possible approaches, and we used a simplified framework that did not compound the concepts upon one another. Our data did not reflect social needs or behavioral health needs, which are increasingly recognized as key facets of complex needs and complex care interventions. Social needs are frequently unavailable as discrete data fields in medical records systems, although some health systems are beginning to systematically collect these data 16,17,18. Our study population was drawn from the membership of a large integrated health system, and thus our results may not be translatable to other patient populations; however, the membership of KPSC is a diverse population that is generally representative of the geography it serves 8 and our method is easily replicated in any patient population.
Our study demonstrates the potential insight that can be gained from early exploration of various specifications for identifying patients with complex needs. Use of data-driven approaches to identify patients with complex needs for program enrollment is a beneficial strategy for a few reasons. First, it bypasses some sources of potential inequity in referral-based program offerings; programs that rely on clinician referral for enrollment have sometimes been criticized because of the potential incursion of implicit biases informing patient selection decisions 19. Second, data-driven methods facilitate the use of rigorous evaluation strategies because it is easier to identify an “administrative control group” for an initial pilot study before the decision has been made to scale an intervention. Lastly, and most directly exemplified by our analysis, data-driven approaches offer the opportunity to explore and compare alternative population definitions and criteria specifications before program launch, which can be informative in choice of population as well as in determining intervention design. Future complex care studies should be intentional and transparent about the design of population selection specifications and report those criteria explicitly.
In complex care interventions, choice of patient population is a critical factor informing the design of program services, staffing, and delivery. The procedures that we used to identify and describe different complex populations for illustrative purposes can be replicated easily in other settings to guide real-world decision-making. It is imperative that program planners consider not only the conceptual approach to population selection but also the implications of the criteria specifications chosen to identify eligible patients. Understanding the baseline needs and characteristics of an intervention population is a potentially high-yield analytical investment that can aid in avoiding pitfalls of complex care pilot program operations and evaluation.
References
Blumenthal D, Anderson G, Burke S, Fulmer T, Jha AK, Long P. Tailoring complex-care management, coordination, and integration for high-need, high-cost patients: A vital direction for health and health care. NAM Perspectives. 2016.
The Better Care Playbook [Available from: https://www.bettercareplaybook.org/.
Long P, Abrams M, Milstein A, Anderson G, Lewis Apton K, Lund Dahlberg M, et al. Effective Care for High-Need Patients: Opportunities for Improving Outcomes, Value, and Health. Washington, DC: National Academy of Medicine; 2017.
Shah T, Lewis C, Tsega M, Abrams M. Quick Reference Guide to Promising Care Models for Patients with Complex Needs. . The Commonwealth Fund; 2019.
Goodwin A, Henschen BL, O’Dwyer LC, Nichols N, O’Leary KJ. Interventions for Frequently Hospitalized Patients and Their Effect on Outcomes: A Systematic Review. Journal of hospital medicine. 2018;13(12):853-9.
DuBard CA, Jackson CT. Active Redesign of a Medicaid Care Management Strategy for Greater Return on Investment: Predicting Impactability. Popul Health Manag. 2018;21(2):102-9.
Davis AC, Osuji TA, Chen J, Lyons LJL, Gould MK. Identifying Populations with Complex Needs: Variation in Approaches Used to Select Complex Patient Populations. Population Health Management. 2020; Epub ahead of print.
Koebnick C, Langer-Gould AM, Gould MK, Chao CR, Iyer RL, Smith N, et al. Sociodemographic characteristics of members of a large, integrated health care system: comparison with US Census Bureau data. Perm J. 2012;16(3):37-41.
Stanton MW, Rutherford M. The high concentration of US health care expenditures: Agency for Healthcare Research and Quality Rockville (MD); 2006.
Joynt KE, Figueroa JF, Beaulieu N, Wild RC, Orav EJ, Jha AK. Segmenting high-cost Medicare patients into potentially actionable cohorts. Healthc (Amst). 2017;5(1-2):62-7.
Locker TE, Baston S, Mason SM, Nicholl J. Defining frequent use of an urban emergency department. Emerg Med J. 2007;24(6):398-401.
Newman RJ, Bikowski R, Nakayama K, Cunningham T, Acker P, Bradshaw D. Outcomes of Embedded Care Management in a Family Medicine Residency Patient-Centered Medical Home. Fam Med. 2017;49(1):46-51.
Tamang S, Milstein A, Sorensen HT, Pedersen L, Mackey L, Betterton JR, et al. Predicting patient ‘cost blooms’ in Denmark: a longitudinal population-based study. BMJ Open. 2017;7(1):e011580.
Fleishman JA, Cohen JW. Using information on clinical conditions to predict high-cost patients. Health Serv Res. 2010;45(2):532-52.
Bates D, Saria S, Ohno-Machado L, Shah A, Escobar G. Big Data In Health Care: Using Analytics To Identify And Manage High-Risk And High-Cost Patients. Health Affairs. 2014;33(7):1123-31.
Sundar KR. Universal screening for social needs in a primary care clinic: a quality improvement approach using the your current life situation survey. The Permanente Journal. 2018;22.
Cantor MN, Thorpe L. Integrating data on social determinants of health into electronic health records. Health Affairs. 2018;37(4):585-90.
LaForge K, Gold R, Cottrell E, Bunce AE, Proser M, Hollombe C, et al. How 6 organizations developed tools and processes for social determinants of health screening in primary care: an overview. The Journal of ambulatory care management. 2018;41(1):2.
Shadmi E, Freund T. Targeting patients for multimorbid care management interventions: the case for equity in high-risk patient identification. International journal for equity in health. 2013;12(1):1-4.
Acknowledgements
All contributors to this manuscript met the criteria for authorship. The authors wish to thank the leaders of the Kaiser Permanente CORAL Initiative and the Kaiser Permanente Care Management Institute’s Care for Complex Needs Program.
Funding
This project was supported by a grant from Kaiser Permanente’s Garfield Memorial Fund, under its Complex Care Collaboration: Research, Operations and Leadership (CORAL) portfolio.
Author information
Authors and Affiliations
Contributions
All authors have contributed to this manuscript in accordance with ICMJE guidelines for authorship. All authors contributed to the study conception and design. Data collection and analysis were performed by Anna Davis and Aiyu Chen. The first draft of the manuscript was written by Anna Davis, Aiyu Chen, Michael Gould, and Thearis Osuji. Subsequent drafts were edited by Michael Gould and John Chen. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Conflict of Interest
The co-authors have no conflicts of interest to disclose.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Prior Presentations
An abstract related to this work was selected for oral presentation at the 2020 AcademyHealth Annual Research Meeting; the abstract was published as part of meeting proceedings although the meeting was moved to a virtual format.
Supplementary Information
ESM 1
(PDF 740 kb)
Rights and permissions
About this article
Cite this article
Davis, A.C., Chen, A., Osuji, T.A. et al. Defining Complex Patient Populations: Implications for Population Size, Composition, Utilization, and Costs. J GEN INTERN MED 37, 351–358 (2022). https://doi.org/10.1007/s11606-021-06815-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11606-021-06815-4