Multimorbidity Patterns in the United States: Implications for Research and Clinical Practice

The increasing prevalence of persons with multimorbidity in many countries has sparked strong growth in research on the epidemiology of multimorbidity, in part to help improve approaches to preventing and managing chronic conditions ( 1–6 ). In this issue of the Journal , Garin and colleagues have made a major contribution to this field of research by examining nationally representative data from studies of noninstitutionalized, predominantly older adults in nine countries that represent the socioeconomic spectrum, and by using a common set of 12 chronic conditions to characterize epidemiologic patterns of multimorbidity among older adults in those countries ( 7 ). Particularly noteworthy are their results for the relation between multimorbidity and sociodemographic factors (age, sex, education, marital status, wealth, and place of residence), as well as the most prevalent comorbid conditions (hypertension, arthritis, and cataract). In addition, their analysis identified selected multimorbidity combinations for each country and across countries, the most common of which are “cardio-respiratory” and “metabolic” patterns. In the United States, more than one-quarter of adults have multiple chronic conditions and the prevalence of multimorbidity has been increasing ( 8 , 9 ). As part of a national-level initiative to help address this challenge, the U.S. Department of Health and Human Services (HHS) developed a “Strategic Framework on Multiple Chronic Conditions” which calls for research to include characterizing the epidemiology and identifying the most common patterns of multimorbidity ( 10–12 ). In this commentary, we examine multimorbidity patterns in the United States and the most common combinations of chronic conditions, as identified by analyses of national-level data sets, and discuss implications these patterns hold for research and clinical care. Like the work of Garin and colleagues, our examination addresses these patterns among noninstitutionalized persons residing in household settings; in addition, however, we report on patterns among representative samples of persons in other settings. We compiled multimorbidity patterns from previously published analyses of HHS data sets in four populations (see Table 1 ): (1) noninstitutionalized persons residing in household settings (National Health Interview Survey [NHIS] for 2010 ( 8 )); (2) persons receiving care in ambulatory settings (National Ambulatory Medical Care Survey [NAMCS] for 2009 ( 13 )); (3) persons receiving care in hospitals (National Inpatient Sample [NIS] for 2009 ( 14 )); and (4) persons receiving care from Medicare in one or more settings such as the home, physicians’ offices, hospitals, and skilled nursing facilities (Centers for Medicare & Medicaid Services [CMS] administrative claims data for Medicare beneficiaries enrolled in the fee-for-service program in 2010 ( 15 , 16 )). Overall Multimorbidity Prevalence by HHS Data Set and Population Sampling Frame, and Number of Conditions in Data Set Overall Multimorbidity Prevalence by HHS Data Set and Population Sampling Frame, and Number of Conditions in Data Set For each data set, overall patterns and the prevalence of combinations were estimated through the use of the same analytic framework and a core set of 20 chronic conditions selected by HHS because they meet the definition for chronicity, are prevalent, and are potentially amenable to public health or clinical interventions or both (The 20 conditions are: arthritis, asthma, autism spectrum disorder, cancer, cardiac arrhythmias, chronic kidney disease, chronic obstructive pulmonary disease, congestive heart failure, coronary artery disease, dementia, depression, diabetes, hepatitis, human immunodeficiency virus infection, hyperlipidemia, hypertension, osteoporosis, schizophrenia, stroke, and substance abuse disorders.) ( 17 ) (see the Supplementary Appendix ). Because the number and type of conditions ascertained by each data set varied and was generally less (ie, 10, 13, 15, and 20) than the 20 core conditions, prevalence estimates for the 20 core conditions and the most frequently occurring dyads and triads of these conditions are not available from all studies. We also used the classification framework developed by Piette and Kerr which includes the categories of concordant conditions (parts of the same overall pathophysiologic risk profile and more likely to be the focus of the same disease and self-management plan) and discordant conditions (not directly related in either their pathogenesis or management) ( 18 ). The overall prevalence of multimorbidity among populations covered by the four data sets varied substantially (range: 26.0% to 71.2%) ( Table 1 ), but followed expected patterns: overall prevalence increased by number of available conditions for analysis, intensity of health care, and age. Applying the Piette and Kerr classification approach ( 18 ) to the NHIS household sample, in all six demographic groups the discordant dyad of hypertension and arthritis was the most prevalent combination, with the highest prevalence (63.0%) in women aged 65 years and older ( Table 2 ). In the ambulatory care sample of the NAMCS data set, the concordant dyad of hypertension and hyperlipidemia was the most prevalent combination in five of the six demographic groups ( Table 2 ). For the NIS sample of hospitalized persons, the potentially concordant dyad of behavioral health conditions—depression and substance abuse—was the most common combination among both men and women aged 18–44 years ( Table 2 ). Most Prevalent Chronic Condition Dyads for U.S. Adults With ≥2 Chronic Conditions, by Survey (National Health Interview Survey [NHIS]*, National Ambulatory Medical Care Survey [NAMCS], † and National Inpatient Sample [NIS] ‡ ), by Age, Sex, and Percent Notes : CHD = Coronary heart disease; COPD = chronic obstructive pulmonary disease. *NHIS: Five most prevalent chronic condition dyads for adults with ≥2 chronic conditions, by sex and age group, percent and (95% confidence interval), 2010 (adapted from Ward (8)). † NAMCS: Physician office visits by patients with the five most prevalent chronic condition dyads for adults with ≥2 conditions, by sex and age group, percent and ( SE ) (when <5 dyads: because fails to meet reliability standard), 2009 (adapted from Ashman (13)). ‡ NIS: Five most prevalent chronic condition dyads for U.S. adults with ≥2 chronic conditions, by sex and age, and (percent), 2009 (adapted from Steiner (14)). Most Prevalent Chronic Condition Dyads for U.S. Adults With ≥2 Chronic Conditions, by Survey (National Health Interview Survey [NHIS]*, National Ambulatory Medical Care Survey [NAMCS], † and National Inpatient Sample [NIS] ‡ ), by Age, Sex, and Percent Notes : CHD = Coronary heart disease; COPD = chronic obstructive pulmonary disease. *NHIS: Five most prevalent chronic condition dyads for adults with ≥2 chronic conditions, by sex and age group, percent and (95% confidence interval), 2010 (adapted from Ward (8)). † NAMCS: Physician office visits by patients with the five most prevalent chronic condition dyads for adults with ≥2 conditions, by sex and age group, percent and ( SE ) (when <5 dyads: because fails to meet reliability standard), 2009 (adapted from Ashman (13)). ‡ NIS: Five most prevalent chronic condition dyads for U.S. adults with ≥2 chronic conditions, by sex and age, and (percent), 2009 (adapted from Steiner (14)). Among the population of Medicare fee-for-service beneficiaries with at least two of the chronic conditions, high cholesterol was the most common condition among the most prevalent dyads: over one-half (52.9%) had the concordant combination of high cholesterol and hypertension, and approximately one third had the concordant dyads with high cholesterol and ischemic heart disease (36.2%) or diabetes (32.3%) ( Table 3 ). Among the costliest dyads, stroke in combination with at least one other condition was the most common co-occurring condition ( Table 4 ). Among the Medicare beneficiary population with at least three of the chronic conditions, each of the five most prevalent triads included high cholesterol in combination with other conditions, including, for example, hypertension and ischemic heart disease (33.7%), hypertension and diabetes (29.9%), and hypertension and arthritis (25.7%) ( Table 3 ) ( 15 ). Top Five Most Prevalent Dyad and Triad Chronic Condition Combinations Among Medicare FFS Beneficiaries With At Least Two Chronic Conditions ( N = 21,437,864) and three Chronic Conditions ( N = 16,481,562), Respectively, by Prevalence and per Capita Medicare Spending, 2010 (adapted from CMS Chartbook (15)) Note : FFS = fee for service. Top Five Most Prevalent Dyad and Triad Chronic Condition Combinations Among Medicare FFS Beneficiaries With At Least Two Chronic Conditions ( N = 21,437,864) and three Chronic Conditions ( N = 16,481,562), Respectively, by Prevalence and per Capita Medicare Spending, 2010 (adapted from CMS Chartbook (15)) Note : FFS = fee for service. Top Five Most Costly Dyad and Triad Chronic Condition Combinations Among Medicare FFS Beneficiaries With At Least Two Chronic Conditions ( N = 21,437,864) and Three Chronic Conditions ( N = 16,481,562), Respectively, by Prevalence and per Capita Medicare Spending, 2010 (adapted from CMS Chartbook (15)) Notes : COPD = chronic obstructive pulmonary disease; FFS = fee for service. Top Five Most Costly Dyad and Triad Chronic Condition Combinations Among Medicare FFS Beneficiaries With At Least Two Chronic Conditions ( N = 21,437,864) and Three Chronic Conditions ( N = 16,481,562), Respectively, by Prevalence and per Capita Medicare Spending, 2010 (adapted from CMS Chartbook (15)) Notes : COPD = chronic obstructive pulmonary disease; FFS = fee for service. Patterns from these analyses have potential implications for clinical decision making and patient management, such as drug–disease, disease–disease, and drug–drug interactions. Examples of these patterns are the high prevalence of hypertension with other conditions across several populations; the presence of both mental and physical health conditions in some subgroups within the NIS population; and, among the Medicare population, the high prevalence of combinations sharing common pathophysiologic risk profiles. The case of hypertension illustrates clinical and preventive implications of a highly prevalent condition when it occurs in combination with certain discordant conditions. Hypertension and arthritis, a discordant dyad, was the most prevalent combination in the NHIS sample of persons in household settings. This combination triggers important drug–disease considerations because of the common use of non-steroidal anti-inflammatory drugs for treating various types of arthritis; these drugs can have both direct effects on blood pressure, and indirect effects through causal pathways that cause renal damage and chronic kidney failure. In addition, diuretics used for treating hypertension may promote hyperuricemia and gout. Secondarily, this combination triggers disease–disease considerations because of the constraints placed by arthritis on physical activity ( 19 ) and, therefore, interference with beneficial effects of exercise on cardiovascular health. In addition to physical health conditions, the HHS strategic framework highlights mental health ( 11 ), and the set of 20 chronic conditions used for the four analyses includes depression, dementia, substance abuse and addiction disorders, autism spectrum disorder, and schizophrenia ( 17 ). Among the NIS hospitalized population, the dyad of depression and substance abuse was the most common combination among both men and women aged 18–44 years ( 14 ). This finding contrasts with patterns for the same age cohort in the NHIS and NAMCS samples reflecting, in part, factors such as the availability of a larger set of conditions for the NIS analysis (ie, all five core mental health conditions in NIS vs none in NHIS and one in NAMCS), and differences in both the populations and settings (ie, hospital vs household, respectively). This finding also underscores the need for more complete information on mental health in data collection efforts. Issues involving the interplay between mental and physical comorbidities have been addressed by Mercer and colleagues and by others, such as how mental health disorders may adversely affect receipt of care for physical conditions, as well as a “bidirectional” relation in which persons with mental health problems are at risk for chronic physical conditions and vice versa ( 20 , 21 ). These interactions between medical and psychological conditions, together with reports on physical comorbidities and neurobehavioral disorders in different populations ( 22–24 ), accentuate the need for detailed study of key combinations of mental and physical conditions. Finally, the Medicare data demonstrate that some conditions that are present in the most prevalent dyads and triads are risk factors for, or precursor conditions to, some of the conditions in the most costly dyads and triads—that is, some of these conditions are in a “causal” relation to others. The Medicare data also show that the high prevalence dyads and triads that include generally asymptomatic conditions, such as high cholesterol or high blood pressure, are characterized by lower per capita spending in the short term (see Table3), although spending for people with the most prevalent dyads or triads still far exceeds average per capita Medicare spending ($9,738) ( 15 ). In the United States, the population burden of multimorbidity is increasing directly as the population ages, with the fastest growth in the segment of adults aged 85 years and older ( 25 ). This trend has exposed a need for new evidence and approaches that support effective clinical management of multimorbidity, including the use of typologies ( 18 , 26 , 27 ), persons with selected index conditions ( 28 , 29 ), and key defined population groups for describing patterns of combinations ( 22 , 30 ). Our findings also contribute to the expanding knowledge base by distilling and exploring implications of recently analyzed, nationally representative data on multimorbidity in a variety of populations and settings. We conclude by outlining how these and other findings can support improvements in the health status of people with multimorbidity by fostering increased use of such data and research findings by clinical practitioners and health systems, and by guiding additional research on combinations of chronic conditions. Keying on specific combinations of conditions and/or symptoms to aid in clinical management is a form of pattern recognition that represents a primary strategy experienced practitioners use to establish a diagnosis, determine a patient’s prognosis, and identify potential treatments ( 31 ). Advances in understanding patterns of and factors underlying combinations of chronic conditions can inform use of pattern recognition, which also was implicitly acknowledged by Garin and colleagues in their discussion of distinct patterns (eg, “cardio-respiratory” and “metabolic”) ( 7 ). Moreover, system supports like electronic health record functionalities (eg, clinical decision supports and identification of drug–drug interactions) might further enhance and systematize approaches that for many clinicians have been informal and intuitive processes. The clinical approach to diabetes is an example of a management strategy that keys on combinations of chronic conditions. Beyond addressing glycemic status, practitioners optimally manage comorbidities that are closely bound to diabetes. This holistic care of related conditions has been codified in quality measures for “optimal diabetes care” ( 32 ). The “discordant” dyad of hypertension and arthritis, so highly prevalent in the NHIS population, is an example of a case in which the electronic medical record and clinical decision support innovations may assist in alerting the provider to potential adverse drug–drug or drug–disease interactions. The many dyads and triads that contain groups of related cardiovascular disease outcomes (eg, heart failure and stroke) and key risk factors (eg, hypertension and hyperlipidemia) represent another area that is ripe for intervention. Hypertension and hyperlipidemia remain significantly under-assessed and under-treated. Many health system interventions (eg, clinical decision support, quality measurement and reporting, and team-based care) have been utilized in high-performing health systems and could help to address these high-priority, high-burden issues ( 33 , 34 ). In addition, interventions aimed at lifestyles and behaviors could address many chronic conditions. For example, interventions that could increase physical activity can be useful in preventing or ameliorating many of the conditions that contribute to the common combinations discussed in this article and by Garin and colleagues, including cardiovascular disease, hyperlipidemia, hypertension, diabetes, osteoarthritis, and depression. Smoking cessation is useful in addressing asthma, chronic obstructive pulmonary disease, cancer, and cardiovascular disease, and interventions to address medication adherence or self-management also might address multiple targets of change. Within the realm of clinical guidelines development, a growing literature is highlighting the importance of considering comorbidities of given guidelines’ target index conditions. Although clinicians are encouraged to follow evidence-based guidelines, guidelines often have been designed for a single chronic condition and are inattentive to issues related to comorbidities ( 35 , 36 ). As a result, recommendations for improving guidelines have noted the importance of, and some organizations have responded to the need for, more explicitly accounting for comorbidities ( 37 , 38 ). Recognition of the importance of comorbidities also highlights the need for identification of specific combinations that may have implications for guidelines for specific conditions. In the longer term, at least four research dimensions on combinations of chronic conditions might be explored. First, basic research on underlying etiologies and identifying common factors underlying multimorbidity may help to improve understanding of pathogenesis and inform intervention development based on mechanisms of disease. Such research might be particularly important for concordant conditions. Second, data for characterizing and better understanding combinations must be substantially expanded by increasing the number of conditions, populations, and settings that are covered in surveys and other data collection efforts. For example, although the HHS data set analyses included in this article used a core set of common conditions and national-level samples, the resulting prevalence estimates nonetheless reflect differences in exclusions from each sample: specifically, NHIS excludes nursing home residents and incarcerated persons; Medicare excludes nondisabled persons younger than 65 years; NAMCS excludes persons who do not seek outpatient care; and NIS excludes persons who are too healthy to require hospital care and persons who do not seek care. Moreover, two data sets—NAMCS and NIS—are based upon the setting in which people are seen (ie, outpatient and inpatient, respectively), with the hospital setting providing the highest intensity of care available for any chronic condition, thus more likely reflecting care for the most severe manifestations of disease. Third, drawing on an expanded body of data and descriptive epidemiology, clinical research involving persons with different combinations of conditions can be directed at identifying risk factors (eg, pathophysiology, care patterns, polypharmacy) that are associated with more severe or rapid clinical deterioration, or with increased likelihood of poor outcomes (eg, readmissions, emergency department visits, and functional impairment). Fourth, further study should be directed at observations arising from combinations in the Medicare population suggesting that the severity of symptoms that are present in conditions composing combinations are principal determinants of cost. Approaches to testing this hypothesis might include examining the degree to which different combinations fit this construct and using findings to assist in clinical management. For example, for selected combinations the fit could be linear or it could abruptly inflect in per capita cost, suggesting combinations that might be priority targets for clinical interventions. An example of research that encompasses these dimensions is NIH-funded clinical outcomes research targeting persons with selected dyads—specifically, the effectiveness and safety of interventions for patients with the highly prevalent dyads of heart failure and chronic lung disease, and heart failure and chronic kidney disease ( 39 ). In addition to observational studies, however, large-scale clinical trials ( 40 ) also are needed to build the knowledge base on clinical outcomes in persons with multimorbidity. Our findings complement those of Garin and colleagues in expanding understanding of the epidemiology of multimorbidity among this population globally. Moreover, the approaches used in both reports provide potential models for additional studies in other countries: both analyses used standard sets of chronic conditions that were applied to nationally representative population samples to produce prevalence estimates of overall multimorbidity and of combinations of conditions. At the same time, each approach had a distinguishing strength: our analysis provided an in-depth examination of multimorbidity in populations in a spectrum of key settings in one country, while that by Garin and colleagues characterized multimorbidity patterns in nine counties with socioeconomic variation. There is a need to identify interventions and outcomes that are for persons with multimorbidity, and health care facilities and systems to those interventions and outcomes across settings of care. research on combinations of chronic conditions can help in approaches for improving of clinical to those persons who are in risk of and of health outcomes the presence of clinical related to this by and was in their as of the and their HHS work on this was not by any or public The are to for of, and on this and the for their of of the and The findings and in this report are those of the and do not represent the of the for and and the Department of Health and Human The in this article do not reflect the of the for Medicare & Medicaid the Health the Department of Health and Human or the United States