1 Mauzahn

Seventy-Five Years Of Comorbidity Research Paper

Psychosocial research

Chronic drinking as a defensive symptom.

Lisansky (1960) asserted, “a basic question that must be posed before considering the problem of personality variables involved in alcoholism is whether alcoholism is a diagnostic entity in and of itself or only a syndrome or symptom” (p. 314). In this regard, Strecker (1941) reflected the prevailing psychoanalytic view on alcoholism of his day in saying that “true chronic alcoholism is a psychoneurosis, defensive in character, with the object of shutting out reality inimical to emotional immaturity—a mechanism which appears to be a logical aftermath of the stunting in childhood of the emotional growth” (p. 14).

This view seems to imply that treating alcoholism directly would be clinically misguided because it is understood to be a mere symptom of a primary psychiatric/psychological condition. However, Tiebout (1951) criticizes the prevailing sentiment that “anyone who stops to study a symptom is thought to be therapeutically naïve and in need of some instruction about first principals” (p. 53). While not disputing that chronic drinking leading to alcoholism could start as a defensive symptom, Tiebout (1951) argued that psychiatrists are mistaken when they conclude, “The drinking is merely a sign of depression, of withdrawal or of some neurotic complication” (p. 53). Instead, he noted that regardless of the initial causes, chronic drinking “finally assumes disease significance and we then treat it as an independent illness” (p. 54). Ten years earlier, Strecker (1941) also concluded that treating the patient’s primary underlying psychological disorder “is impossible so long as he uses alcohol” (p. 14). Notably, the disease versus symptom question raised by Lisansky above, and still with us today, has diminishing practical importance if both conditions require specific treatment regardless.

The pharmacobehavioral tension-reduction theory (PB-TRT).

While the empirical search for a particular psychoanalytic personality disturbance (e.g., “latent homosexuality”) that predisposes one toward alcoholism consistently failed (e.g., Bowman and Jellinek, 1941b), the idea that pathological drinking was used as a buffer against distressing psychological feelings continued to thrive. For example, Ullman (1952) suggested that chronic drinking can be “a tension-reducing activity with the source of tension lying in the ordinary problems of human beings” (p. 603). By evoking Thorndike’s “law of effect,” Ullman argued that when drinking relieves distress (even when that distress is caused by “ordinary” problems), it would be reinforced and thereby escalate, potentially to the level of an independent disorder. “What this danger consists of is the possibility that the response may become so exaggerated that it becomes a problem or even a disease itself” (Ullman, 1952, p. 604).

Conger (1956) continued to develop the PB-TRT by embedding it in a learning-based drive-reduction model and using a variety of experimental analog models with animal subjects to test specific hypotheses derived from the psychoanalytic view of alcoholism. For example, he cited several studies showing that experimentally induced “neurosis”—indicated when the animal decreases the frequency of making a response to receive food once that response is also paired with a shock—is reversed by alcohol (i.e., the animal increases responding for food again). “Before alcohol the avoidance is stronger than approach; after alcohol the avoidance is weaker than approach” (Conger, 1956, p. 299). However, Conger continued to understand this work as a mere analog to the real world of clinical dysfunction in which it remained important to find out “. . . what need or drive patterns are particularly important among various kinds of alcoholics. The work of the psychoanalyst in this regard is a case in point” (p. 304). Up to this time, then, the conceptual continuity between the TRT and the psychoanalytic view of alcoholism remained explicit.

However, by the early 1970s, many dozens of laboratory studies had tested the PB-TRT using various learning paradigms with little or no reference to psychoanalytic precepts; that is, the TRT had evolved from a laboratory-based psychoanalytic analog model into a distinct scientific species. Cappell and Herman (1972) reviewed this literature decomposing the PB-TRT into two hierarchal hypotheses: (a) that alcohol reduces tension and (b) that alcohol is consumed for its tension-reducing properties. Importantly, their review was concerned only with experiments testing the former hypothesis because “clearly it makes little sense to assert that tension relief motivates drinking unless it can be demonstrated that such relief is indeed a consequence of alcohol administration” (Cappell and Herman, 1972, p. 34). They concluded that the TRT (a) was not supported by studies using avoidance and escape paradigms, (b) was supported by studies using conflict and experimental neurosis paradigms, and (c) was understudied in the area of subjective effects of alcohol on negative emotions and frank psychiatric symptoms in humans. Their overall conclusion regarding the PB-TRT, however, was unmistakably pessimistic: “Much of the evidence is negative, equivocal and often contradictory” (p. 33).

The cognitive-behavioral tension-reduction theory (CB-TRT).

An early manifestation of the cognitive revolution described above can be seen in the “cognitive social-learning” revision to the TRT outlined by Donovan and Marlatt (1980). Whereas the PB-TRT focused on alcohol’s pharmacologically mediated tension-reducing effects, the CB-TRT focused primarily on the subjective appraisals and beliefs affecting the likelihood that alcohol would be used to cope with distress. In fact, the CB-TRT does not technically require that alcohol be tension reducing at all, only that one expects this effect in the context of other relevant conditions: “. . . an individual’s cognitive expectancies concerning the effects of alcohol may exert a greater degree of control over drinking and subsequent behavior than the pharmacological effects of the drug” (p. 1159). Conceptually, this view stands in direct contrast to Cappell and Herman’s (1972) assertion that the TRT model of drinking is viable only insofar as alcohol can reliably reduce tension.

As predicted by the CB-TRT, studies have shown that individuals consume more alcohol than usual when drinking for the purpose of coping with distress and that drinking for this purpose is most likely among those who (a) expect alcohol to be effective in reducing distress and (b) doubt their own effectiveness in managing distress without alcohol (Jung, 1977; Kassel et al., 2000; Laurent et al., 1997). (Predictions from CB-TRT studies are also supported for marijuana use; Johnson et al., 2009). Findings showing that those with alcohol problems are more likely than others to expect tension reduction from alcohol and to drink to obtain these effects add face validity to the CB-TRT as a psychosocial model for the etiology of alcohol dependence (Beckman, 1980; Carpenter and Hasin, 1999; Russell and Bond, 1980).

Status of the TRT as a comorbidity model.

These findings, although consistent with CB-TRT, do not clarify the extent to which it is a general etiological model of substance use disorders (all individuals), a general model of comorbidity (all individuals with any mental disorder), or a specific model of comorbidity (only individuals with a specific type[s] of mental disorder). In fact, since the TRT split off from its psychoanalytic progenitor, the theory has been conspicuously agnostic concerning the quality and quantity of distress most relevant to the model, with many studies purporting to test the TRT operationalizing stress without any reference to mental illness (Linsky et al., 1985). Nonetheless, numerous studies do show that mental conditions characterized by intrapsychic distress are robust correlates of drinking to cope, including trait anxiety (Brown and Munson, 1987), depression and fear (Hussong et al., 2005; Martens et al., 2008), social anxiety (Tran et al., 2004), posttraumatic stress disorder (S. E. Ullman et al., 2005), anxiety sensitivity (O’Connor et al., 2008), personality pathology (Bruce et al., 2013), and generalized anxiety (Litt et al., 2013). The breadth of these findings, along with those showing that significant negative life transitions (e.g., job loss) and chronic stress (e.g., poverty) are associated with drinking, suggest that the TRT is not specific to one type of mental disorder and may suggest, as did Ullman (1952), that various forms of mental illness are simply one possible source of chronic stress that can motivate alcohol or other drug use as a means of coping. Further, Koob (e.g., Koob, 2013) has argued persuasively that drinking itself (i.e., its effect on the brain’s stress-response systems) can become one source of negative affectivity that drives TRT processes in substance dependence.

A final issue relevant to the TRT as a comorbidity model stems from the reliable finding that drinking to cope is associated with alcohol-related problems over and above the amount of alcohol being used (Cutter and O’Farrell, 1984; Kassel, et al., 2000; Perkins, 1999). These findings seem anomalous to the supposition that escalating alcohol use via negative reinforcement is the process by which drinking to cope confers risk for the development of substance use disorders and may implicate drinking to cope as an endophenotypic marker or prodromal status related to risk for the development of substance use disorders (e.g., Kushner et al., 2011, 2012a; Menary et al., 2011).

Biomedical comorbidity research

Symptom versus disease conundrum.

Bowman and Jellinek (1941a) observed the “modern standpoint” to be “that the connection between heavy drinking and many of the so-called alcoholic mental disorders is not causative but that the drinking itself is the symptom of a psychosis” (p. 312). (Again, psychosis at the time of their writing referred to mental and behavioral disturbances that did not necessarily include formal thought disorders.) However, they were at pains to add, “The term ‘modern standpoint’ is not intended to imply that this is a discovery of recent date, but to denote the contemporaneous adoption of this viewpoint by a majority” (pp. 312–313). They concluded that “To venture on distinguishing between drinking merely incidental to the psychosis and drinking as a precipitating factor of the nonalcoholic psychosis, although desirable, is not altogether feasible at the present stage of knowledge” (p. 316). Lewis (1941) also struck a skeptical position regarding the assertion by prominent investigators of his day that the general problem of “alcoholic psychosis” is a result of years of heavy drinking: “I can place very little confidence in the statistics on alcoholic psychosis as at best these statistics are recording the fact, and only the fact, that an excessive use of alcohol was somewhere in the picture. However, in many instances, it may have had nothing to do with the etiology or even with the hospitalization of the patient” (p. 295).

Efforts at overcoming taxonomic and methodological chaos in the pre-modern comorbidity research.

It was clear in the pre-modern era that research was needed to quantify comorbidity and discover the etiological relations among comorbid disorders. However, without the methods for making reproducible diagnostic decisions concerning the presence or absence of specific mental syndromes, this agenda was severely restricted. May and Ebaugh (1953), for example, concluded that terms such as alcoholic hallucinosis and psychosis are “obsolete and inadequate” (p. 226). Because these diagnostic categories were defined idiosyncratically by different investigators, the scientific value of knowing how many cases of “alcoholic psychosis” occurred in New York State in 1940 (Malzberg, 1944) or in Massachusetts between 1917 and 1933 (Dayton et al., 1942) was very low indeed.

Recognizing these problems, Freed (1970) laid out a well-developed and ambitious agenda for the study of comorbidity that anticipated the neo-Kraepelinian zeitgeist that was soon to overtake psychiatry. He called for establishing reliable diagnostic criteria as a prerequisite to the serious study of comorbidity and recommended (a) epidemiological studies to quantify the actual rate at which well-defined psychiatric disorders co-occur with alcoholism, (b) longitudinal studies to determine the temporal order of onset and also the symptom covariations of comorbid disorders, and (c) family transmission studies to evaluate the possible shared neurobiological, genetic, or psychiatric “kinship” of comorbid disorders. Notably, these suggestions closely parallel those made independently by Feinstein (1970) in his seminal paper on the conceptual and practical implications of medical comorbidities. With the introduction of the Research Diagnostic Criteria (RDC) and the Diagnostic and Statistical Manual of Mental Disorders, Third Edition (DSM-III; American Psychiatric Association, 1980), both based to a large extent on the earlier Feighner criteria (Feighner et al., 1972), the neo-Kraepelinian research program finally had the tools necessary for its implementation.

Modern epidemiological studies of comorbidity

Weissman and colleagues (1980) were among the first to produce scientifically valid psychiatric epidemiological data concerning comorbidity using (a) a well-specified and agreed upon set of diagnostic criteria, (b) a structured diagnostic interview delivered by trained interviewers, (c) tests of inter-rater reliability among interviewers, and (d) established protocols for settling diagnostic disagreements. Their results showed that an astounding 70% of individuals who were diagnosed as having a lifetime diagnosis of an RDC alcohol use disorder were also diagnosed as having had at least one other lifetime RDC psychiatric disorder, the most common of which was major depression (44%) followed by “minor depressive disorders” and generalized anxiety disorder.

While the methods and findings of the Weissman study were groundbreaking, it was regionally restricted. Helzer and Pryzbeck (1988) used the Epidemiological Catchment Area survey, including about 20,000 individuals sampled from five widely distributed urban centers and their associated rural areas in the United States, to provide the first unbiased estimates of how the odds of being diagnosed with several common DSM-III mental disorders were modified when an alcohol disorder was present. Their results showed that having a lifetime history of alcohol disorder increased the odds of having a lifetime history of antisocial personality disorder by 21-fold; mania by 6-fold; schizophrenia by 4-fold; panic disorder by 2-fold; obsessive–compulsive disorder, major depression, or dysthymia by 2-fold; and any one of the common phobic disorders by 1.4-fold.

The Helzer and Pryzbeck (1988) data also highlighted complexities in interpreting epidemiological data that are affected by how the comorbid associations are characterized and by moderating variables such as gender. For example, although phobias had the least increase in the odds of being diagnosed when an alcohol disorder is present and mania had one of the highest, these positions were reversed when cast in terms of the absolute number of comorbid cases (i.e., phobia was the most common psychiatric disorder co-occurring with alcohol disorder, whereas mania was the least common). Also, the increase in risk for antisocial personality disorder when alcohol disorder was present was four times greater among women compared with men; however, once again, the absolute number of cases of antisocial personality disorder was substantially higher in men.

Modern longitudinal studies of comorbidity

Study of the temporal patterns of comorbid disorder— what Feinstein (1970) referred to as chronometry—is an intuitively appealing method for inferring causal relationships based on the simple logical proposition that only an earlier event can cause a later event. (However, post hoc ergo propter hoc—assuming that an earlier event must have caused a later event simply because it came first—is a logical error that is common in interpreting longitudinal data.) While longitudinal studies are typically focused on the temporal priority of comorbid disorder onsets, additional parameters of interest are the comorbid disorders’ temporal (a) proximity of symptoms, (b) priority of offsets/remissions, and (c) relationship of relapses.

Temporal priority.

Respondents in the Weissman et al. (1980) study reported that their depression began before their alcohol disorder in 60% of the cases in which both conditions were present. Notwithstanding caveats related to the retrospective method used, these findings do imply that the alcohol disorder did not cause comorbid depression in a majority of cases; however, this does not demonstrate that either disorder caused the other. A study by Sexton et al. (1999) also suggested that gender is an important moderator in these temporal associations. Over the 7-year interval in which they followed 8,000 individuals, earlier drinking tended to predict later worsening of depression in men but predicted improvement of depression in women. By contrast, Repetto et al. (2004) reported that higher depression in high school predicted greater future alcohol use in boys but not girls. Reminding us again that amount of substance use is not synonymous with problems caused by the use, Mason et al. (2008) reported that after controlling for gender and depressed mood, alcohol problems but not amount of alcohol use predicted later major depressive episodes in adolescents. Also reminding us that stressful experiences are not synonymous with stressful reactions to those situations, Wu et al. (2010) reported that alcohol use initiation was more common within 2 years of a trauma in 10- to 13-year-olds, but only among those who developed significant symptoms of posttraumatic stress disorder. Although these data seem to confuse more than clarify the causality question, they do suggest that the likelihood of finding a single unidirectional causal pattern in comorbid disorder onset—something of an idée fixe among many comorbidity-focused clinicians and researchers—is rather low.

Temporal proximity.

Schuckit and colleagues (2013) considered the temporal proximity of the manifestations of comorbid disorders by prospectively evaluating “depressive episodes” that either occurred during episodes of heavy drinking (“substance-induced”) or outside of periods of heavy drinking (“independent”). They found that approximately 15% of the sample had developed an independent depressive episode by the conclusion of the 30-year follow-up. They also found that about 30% of the depressive episodes experienced by those who developed an alcohol use disorder diagnosis occurred in the context of periods of heavy drinking. This is fairly consistent with the finding by Weissman et al. (1980) that alcoholism began before depression for about 40% of those with both disorders. Of particular importance was their finding that those with independent depressive episodes had “no increased rate of AUDs [alcohol use disorders] and evidenced no higher rate of use or abuse/ dependence on illicit substances” (Schuckit et al., 2013, p. 271). These findings provide compelling support for the importance of distinguishing subgroups of individuals with depression (and likely other psychiatric disorders) whose symptoms do versus do not become intermingled with substance use.

Disorder remissions (“offsets “).

Another common logical error is to assume that removal of the initial cause of a disorder would necessarily resolve that disorder; that is, initiating and maintaining causes can be distinct. For example, in a 40-year follow-up study of more than 200 males at high risk for alcoholism, Penick et al. (2010) observed that there is “a striking disconnect between measures that predicted alcohol dependence and measures that predicted remission from alcohol dependence” (p. 215). This suggests that successfully treating the causal disorder in a comorbid pair would not necessarily resolve the other condition. With that said, it does appear that the elevated levels of anxiety and depression symptoms that are evident in many individuals at the beginning of an inpatient alcoholism treatment decline as they accommodate to their new environment, emerge from acute withdrawal, and otherwise respond to medical and psychosocial treatments while remaining abstinent (Brown and Schuckit, 1988; Schuckit et al., 1990). However, this does not clarify whether individuals meeting diagnostic criteria for specific psychiatric disorders while actively misusing drugs or alcohol would continue to meet these criteria after a period of abstinence.

Penick et al. (1988), for example, provided data showing that the absolute number and relative risk of psychiatric syndromes identified in substance use disorder treatment patients were stable up to 1 year following treatment; however, it is not clear how continued sobriety versus relapse figured into these outcomes. Verheul et al. (2000) found that recovery of substance use covaries with recovery from some but not other types of psychiatric problems. Ramsey et al. (2004) found that up to one third of patients deemed to have “secondary” depression continued to be clinically depressed up to 1 year after successful alcoholism treatment. These findings are consistent with other epidemiological data from the National Epidemiologic Survey on Alcohol and Related Conditions showing that “only a few individuals” had anxiety or mood disorders that both began after the onset of an alcohol disorder and ceased being symptomatic during periods of prolonged abstinence (Grant et al., 2004, p. 107). Similarly, it appears that successful treatment of an anxiety disorder does not appreciably affect comorbid hazardous drinking and AUDs (Thomas et al., 2008). These data would appear to suggest that even if a psychiatric disorder or substance use disorder is caused (“induced”) by the other, it would not necessarily resolve once the primary condition was successfully treated. This conclusion is also consistent with the observations and intuitions of several pre-modern researchers that secondary comorbid conditions can evolve into independent disorders (e.g., Ullman, 1952).

Relapse.

A final issue to consider in distinguishing initiating from maintaining causal influences in comorbid disorders is the relative “no man’s land” of relapses to substance use in this formulation. Perhaps causal influences related to initiating disorder onsets, but not their offsets, regain causal potency in terms of relapse. For example, relapses to alcohol and drug use following treatment are commonly associated with exacerbations in psychiatric symptoms, and patients often attribute their relapses to worsening affect and anxiety (Kranzler et al., 1996; Najavits et al., 2007); however, disentangling the cause-and-effect relationships between comorbid conditions and relapse remains notoriously difficult (Booth et al., 1991). For example, both affective disorders and substance use disorders are characterized by cycles of remission and relapse even when no comorbidity is present.

Modern family studies of comorbidity

Family studies probe the possibility of a shared heritable etiology of comorbid disorders by evaluating their direct (“true”) transmission (e.g., depression in proband to depression in offspring) versus their cross-transmission (e.g., depression alone in proband to alcoholism alone in offspring). Merikangas et al. (1985) used this approach in comparing rates of depression and alcoholism in the offspring of probands with either depression and “secondary” alcoholism or depression alone. The general conclusion reached by the authors was “that depression and alcoholism are not alternate forms of expression of the same underlying illness” (p. 199). However, they also noted, “some evidence suggested that specificity of transmission existed for the combination and the order of presentation of the two disorders among the adult first-degree relatives” (p. 202). These findings and conclusions are also largely consistent with those reported by Dawson and Grant (1998). This is quite provocative in suggesting that the combination of alcoholism and depression (and perhaps other mental disorders) may constitute a genetically distinct disorder subtype.

At least two family studies have failed to find evidence of cross-transmission as an explanation for comorbidity between anxiety disorder and alcohol use disorder. Schuckit et al. (1995) reported that after controlling for substance-induced “organic” anxiety syndromes and controlling for assortative mating in the parents of probands, there was no evidence of an exceptionally high rate of anxiety disorders in the close relatives of alcoholic men and women. Similarly, Blonigen et al. (2013) found that the within-twin-pair differences in lifetime internalizing disorders (comprised of common affective and anxiety disorders) were significantly related to the within-twin-pair differences in the development of alcohol problems 10 years later (i.e., the twin with the earlier internalizing disorder had more risk for the development of later alcohol problems). They concluded that a history of such psychiatric problems “appears to be linked to problem drinking in midlife above and beyond the confounding influence of genetic effects” (p. 136).

In a small family history study, Winokur et al. (1993) found that, among individuals with bipolar disorder, the rate of alcoholism in their family members was not statistically different between the individuals who did (21%) versus did not (35%) have a co-occurring alcohol disorder. Given the low number of participants and the absence of control groups, it is hard to interpret these findings, but they are consistent with a cross-transmission pattern indicating a shared genetic liability to both disorders. This is consistent with other findings pointing to the possibility of a specific genetic parallel across mania and alcohol use disorders (Schuckit et al., 2003).

Abstract

Multimorbidity, the coexistence of 2 or more chronic conditions, has become prevalent among older adults as mortality rates have declined and the population has aged. We examined population-based administrative claims data indicating specific health service delivery to nearly 31 million Medicare fee-for-service beneficiaries for 15 prevalent chronic conditions. A total of 67% had multimorbidity, which increased with age, from 50% for persons under age 65 years to 62% for those aged 65–74 years and 81.5% for those aged ≥85 years. A systematic review identified 16 other prevalence studies conducted in community samples that included older adults, with median prevalence of 63% and a mode of 67%. Prevalence differences between studies are probably due to methodological biases; no studies were comparable. Key methodological issues arise from elements of the case definition, including type and number of chronic conditions included, ascertainment methods, and source population. Standardized methods for measuring multimorbidity are needed to enable public health surveillance and prevention. Multimorbidity is associated with elevated risk of death, disability, poor functional status, poor quality of life, and adverse drug events. Additional research is needed to develop an understanding of causal pathways and to further develop and test potential clinical and population interventions targeting multimorbidity.

aged, chronic disease, comorbidity, prevalence

INTRODUCTION

Multimorbidity, the coexistence of 2 or more chronic conditions, has become widely prevalent through the third phase of the epidemiologic transition, which is characterized by a decline in mortality rates combined with an aging population (1). Recognizing the importance of multiple chronic conditions, primary care practitioners have adopted a patient-centered focus on multimorbidity, and researchers are increasingly interested in understanding the phenomenon. In developed countries, the prevalence of multimorbidity in the older population and its impact on health-care expenditures have led health agencies to begin to address the problem and explore ways to improve health and function (2).

Comorbidity, the conceptual predecessor of multimorbidity, was originally defined by Feinstein as “any distinct additional clinical entity that has existed or may occur during the course of a patient who has the index disease under study” (3, pp. 456–457). This initial disease-centered approach to research might have led to a predominant focus on the uncomplicated “index” disease and resulted in a paucity of information about the complex and all-too-common multimorbid patient.

The purpose of the present review was to examine several questions related to multimorbidity and comorbidity: 1) What is the prevalence of multimorbidity in older adults, particularly those living in the community? 2) How does multimorbidity affect health outcomes? 3) What are the implications of multimorbidity for public health and medicine? We examined population-based data on an extremely large sample of US Medicare enrollees and conducted a systematic literature review to address the first question, and we conducted a survey of the literature to address the second and third questions.

MATERIALS AND METHODS

We defined multimorbidity as the presence of 2 or more chronic conditions, consistent with the US Department of Health and Human Services framework (4). We limited prevalence studies to those reporting results for community samples, and we excluded samples sourced solely from care settings or limited to younger age groups.

Medicare population

The Centers for Medicare and Medicaid Services developed a database of administrative claims data for 100% of Medicare beneficiaries who were continuously enrolled in fee-for-service coverage in Medicare Parts A and B for the entire year of 2008. The presence of each of 15 chronic conditions (listed in Table 1) was identified through claims data on the basis of evidence of treatment or service delivery for each condition. “Cancer” included breast, colon, lung, and prostate cancer. A complete description of the methodology, including attribution of chronic conditions, can be found at the Chronic Condition Data Warehouse website (5). The number of chronic conditions was counted for each person and grouped in various ways for analysis. Figures and tables were adapted from the summary data for the Centers for Medicare and Medicaid Services’ Chronic Conditions Among Medicare Beneficiaries, Chartbook, 2011 Edition (6) or from summary analysis (Kimberly Lochner, Centers for Medicare and Medicaid Services, personal communication, 2012).

Table 1.

Percentage of Medicare Beneficiaries With Selected Chronic Conditions, by Age and the Presence of Comorbidity, United States, 2008a

Chronic Condition Prevalence, %
% With Comorbidity 
Overall Age ≥65 Years 
Hypertension 56.2 59.6 93.5 
Hyperlipidemia 42.8 45.4 94.9 
Ischemic heart disease 32.0 34.5 96.1 
Diabetes 26.6 26.9 95.1 
Arthritis 20.8 22.2 93.6 
Heart failure 16.8 18.0 98.7 
Depression 13.1 10.7 90.0 
Chronic kidney disease 12.7 13.1 98.1 
Osteoporosis 12.4 13.9 92.4 
Alzheimer's disease 11.0 12.6 94.0 
Chronic obstructive pulmonary disease 10.9 11.1 96.6 
Atrial fibrillation 7.7 8.9 97.9 
Cancerb6.5 7.4 91.3 
Asthma 4.5 4.0 95.4 
Stroke 4.3 4.6 98.5 
Chronic Condition Prevalence, %
% With Comorbidity 
Overall Age ≥65 Years 
Hypertension 56.2 59.6 93.5 
Hyperlipidemia 42.8 45.4 94.9 
Ischemic heart disease 32.0 34.5 96.1 
Diabetes 26.6 26.9 95.1 
Arthritis 20.8 22.2 93.6 
Heart failure 16.8 18.0 98.7 
Depression 13.1 10.7 90.0 
Chronic kidney disease 12.7 13.1 98.1 
Osteoporosis 12.4 13.9 92.4 
Alzheimer's disease 11.0 12.6 94.0 
Chronic obstructive pulmonary disease 10.9 11.1 96.6 
Atrial fibrillation 7.7 8.9 97.9 
Cancerb6.5 7.4 91.3 
Asthma 4.5 4.0 95.4 
Stroke 4.3 4.6 98.5 

View Large

Systematic review

We searched for articles that described the prevalence of multimorbidity in studies conducted in the general population. Using Medical Subject Headings and keywords, we conducted an electronic literature search of the PubMed database for English-language articles published between 1980 and May 2012. The complete search (shown in the Appendix) was conducted, and then articles on nonhuman and nonelderly studies were excluded. We included some papers identified through manual searching and some citations from other reviews and those recommended by selected experts, including unpublished manuscripts. We reviewed the abstracts to exclude articles that were not eligible. This review excluded studies conducted only in health-care settings, such as primary care offices or inpatient hospitals; studies without older adults aged ≥65 years; and studies that did not report the overall or age-specific prevalence of multimorbidity. Studies with sample sizes under 500 were excluded as well. Articles were not subjected to quality assessment. We reviewed the full text of retrieved papers. We extracted data on prevalence by age group for groupings above 59 years from articles that met all inclusion criteria.

RESULTS

Medicare population

A total of 30,923,846 persons were enrolled in Medicare fee-for-service continuously during 2008, of whom 16.5% were under 65 years of age and were eligible because of disability or end-stage renal disease. The most prevalent chronic conditions were hypertension, hyperlipidemia, and ischemic heart disease (Table 1). The least prevalent chronic conditions were atrial fibrillation, cancer, asthma, and stroke, each occurring in less than 10% of the population. Among persons with the 15 “index” conditions of interest, the vast majority had at least 1 other comorbid condition, ranging from 90% for those with depression to 98.7% for those with heart failure.

In 2008, 33% of Medicare beneficiaries had 0 or 1 chronic condition, whereas 67% had multimorbidity (2 or more chronic conditions), and the prevalence of multiple chronic conditions increased with age (Figure 1). The prevalences of 4, 5, and 6 or more chronic conditions increased with age, and this was most pronounced for 6 or more conditions (Figure 1). The most prevalent combination of 2 chronic conditions was hypertension and hyperlipidemia, and the most prevalent combination of 3 conditions was hypertension, hyperlipidemia, and ischemic heart disease; both combinations would be predicted from the individual prevalence rates.

Figure 1.

Percentage of the US population enrolled in the Medicare program, by number of chronic conditions and age group, 2008. Data were obtained from Centers for Medicare and Medicaid Services (CMS) administrative claims data, January–December 2008, accessed from the CMS Chronic Condition Data Warehouse (5). The graph was adapted from Chronic Conditions Among Medicare Beneficiaries, Chartbook, 2011 Edition (6).

Figure 1.

Percentage of the US population enrolled in the Medicare program, by number of chronic conditions and age group, 2008. Data were obtained from Centers for Medicare and Medicaid Services (CMS) administrative claims data, January–December 2008, accessed from the CMS Chronic Condition Data Warehouse (5). The graph was adapted from Chronic Conditions Among Medicare Beneficiaries, Chartbook, 2011 Edition (6).

In 2008, 67% of Medicare beneficiaries had multimorbidity, and its prevalence increased with age, from 62% between 65 and 74 years of age to 81.5% at ≥85 years of age (Table 2). Within each age group, women had a higher prevalence of multimorbidity than men, most prominently in the youngest age group and less so above age 85 years.

Table 2.

Percentage of Medicare Beneficiaries With Multimorbidity (≥2 of 15 Selected Chronic Conditionsa), by Age and Gender, United States, 2008b

Age, years Prevalence, %
Men Women Overall 
<65 45.7 55.4 50.3 
65–74 59.9 63.9 62.0 
75–84 73.4 77.4 75.7 
≥85 79.5 82.3 81.5 
Age, years Prevalence, %
Men Women Overall 
<65 45.7 55.4 50.3 
65–74 59.9 63.9 62.0 
75–84 73.4 77.4 75.7 
≥85 79.5 82.3 81.5 

View Large

Prevalence of multimorbidity

Through the article selection process (Figure 2), we identified 17 studies on the prevalence of multimorbidity. The final sample of prevalence studies contained national rates reported from the United States, Australia, Canada, Ireland, Israel, and Spain and regional or local rates from many European nations (Table 3) (7–22). The sample sizes of studies ranged from approximately 1,000 to nearly 31 million, with the largest sample sizes being from Medicare claims databases. Study methods included the use of national samples, claims databases, and recruited geographic cohorts. Most studies relied on a self-reported diagnosis from a health professional or used diagnostic codes or use of medications from administrative claims data, but a few used direct clinical assessments or mixed methods. The number of chronic conditions considered ranged from 7 to more than 30.

Table 3.

Prevalence of Multimorbidity in the General Population and Study Sample and Related Characteristics

First Author, Year (Reference No.) Country No. of Persons Age, years Data Source No. of Conditions Considered Prevalence of Multimorbidity, % 
Verbrugge, 1989 (7) United States 16,148 ≥55 National sample survey; self-report 13 63.1 
Hoffman, 1996 (8) United States 34,459 All National sample survey; self-report All chronic conditions classified by ICD-9 codes 69 (age ≥65 years) 
Fuchs, 1998 (9) Israel 1,487 75–94 Community survey; self-report 14 64.5 
Menotti, 2001 (10) Finland, the Netherlands, Italy 716 (Finland), 887 (the Netherlands), 682 (Italy) Men 65–84 Geographically recruited cohorts; clinical examination 23.3 (Finland), 13.1 (the Netherlands), 15.3 (Italy) 
Wolff, 2002 (11) United States 1,217,103 ≥65 Medicare claims data; sample 23 groups 65 
Partnership for Solutions, 2004 (12) United States NR All National sample survey; self-report All chronic conditions classified by ICD-9 codes 67 (age ≥65 years) 
Rapoport, 2004 (13) Canada 17,244 >20 National sample survey; self-report 22 54.7 (age 60–79 years), 64 (age ≥80 years) 
Naughton, 2006 (14) Ireland 316,928 ≥70 National pharmacy claims database; drug dispensing 60.4 
Broemeling, 2008 (15) Canada NR ≥12 National sample survey; self-report 35 (age 60–79 years), 48 (age ≥80 years) 
Britt, 2008 (16) Australia 9,156 All National sample of 305 general practitioners 8 domains + cancer 75 (age 65–74 years), 83 (age ≥75 years) 
Nagel, 2008 (17) Germany 13,781 50–75 Geographically recruited cohort; self-report 13 groups 67.3 
Marengoni, 2008 (18) Sweden 1,099 77–100 Geographically recruited cohort; clinical assessment 30 55 
Schram, 2008 (19) The Netherlands 2,463 (LASA), 3,550 (Rotterdam), 599 (Leiden) 55–94 (LASA), ≥65 (Rotterdam), 85 (Leiden) Geographically recruited cohort; self-report and clinical examination, varying by site 12 (LASA), 15 (Rotterdam), 13 (Leiden) 56 (LASA), 72 (Rotterdam), 67 (Leiden) 
Schneider, 2009 (20) United States 1,649,574 All Medicare claims data; sample 20 
Loza, 2009 (21) Spain 2,192 >20 National sample survey; self-report All diseases 30 
Centers for Medicare and Medicaid Services, 2011 (present report) United States 30,923,846 All Medicare claims data; all fee-for-service 15 62 (age 65–74 years), 76 (age 75–84 years), 81 (age ≥85 years) 
Kirchberger, 2012 (22) Germany 4,127 65–94 Geographically recruited cohort; self-report 13 58.6 
First Author, Year (Reference No.) Country No. of Persons Age, years Data Source No. of Conditions Considered Prevalence of Multimorbidity, % 
Verbrugge, 1989 (7) United States 16,148 ≥55 National sample survey; self-report 13 63.1 
Hoffman, 1996 (8) United States 34,459 All National sample survey; self-report All chronic conditions classified by ICD-9 codes 69 (age ≥65 years) 
Fuchs, 1998 (9) Israel 1,487 75–94 Community survey; self-report 14 64.5 
Menotti, 2001 (10) Finland, the Netherlands, Italy 716 (Finland), 887 (the Netherlands), 682 (Italy) Men 65–84 Geographically recruited cohorts; clinical examination 23.3 (Finland), 13.1 (the Netherlands), 15.3 (Italy) 
Wolff, 2002 (11) United States 1,217,103 ≥65 Medicare claims data; sample 23 groups 65 
Partnership for Solutions, 2004 (12) United States NR All National sample survey; self-report All chronic conditions classified by ICD-9 codes 67 (age ≥65 years) 
Rapoport, 2004 (13) Canada 17,244 >20 National sample survey; self-report 22 54.7 (age 60–79 years), 64 (age ≥80 years) 
Naughton, 2006 (14) Ireland 316,928 ≥70 National pharmacy claims database; drug dispensing 60.4 
Broemeling, 2008 (15) Canada NR ≥12 National sample survey; self-report 35 (age 60–79 years), 48 (age ≥80 years) 
Britt, 2008 (16) Australia 9,156 All National sample of 305 general practitioners 8 domains + cancer 75 (age 65–74 years), 83 (age ≥75 years) 
Nagel, 2008 (17) Germany 13,781 50–75 Geographically recruited cohort; self-report 13 groups 67.3 
Marengoni, 2008 (18) Sweden 1,099 77–100 Geographically recruited cohort; clinical assessment 30 55 
Schram, 2008 (19) The Netherlands 2,463 (LASA), 3,550 (Rotterdam), 599 (Leiden) 55–94 (LASA), ≥65 (Rotterdam), 85 (Leiden) Geographically recruited cohort; self-report and clinical examination, varying by site 12 (LASA), 15 (Rotterdam), 13 (Leiden) 56 (LASA), 72 (Rotterdam), 67 (Leiden) 
Schneider, 2009 (20) United States 1,649,574 All Medicare claims data; sample 20 
Loza, 2009 (21) Spain 2,192 >20 National sample survey; self-report All diseases 30 
Centers for Medicare and Medicaid Services, 2011 (present report) United States 30,923,846 All Medicare claims data; all fee-for-service 15 62 (age 65–74 years), 76 (age 75–84 years), 81 (age ≥85 years) 
Kirchberger, 2012 (22) Germany 4,127 65–94 Geographically recruited cohort; self-report 13 58.6 

View Large

Figure 2.

Number of references identified at each stage of a systematic review of multimorbidity among older adults.

Figure 2.

Number of references identified at each stage of a systematic review of multimorbidity among older adults.

Age-specific and overall prevalence rates extracted from the articles are summarized in Table 3. The prevalence of multimorbidity in the reviewed studies ranged from 13% (10) to 83% (age ≥75 years) (16), with a median of 63% and a mode of 67%. The prevalence rates were lower for studies that included fewer than 10 chronic conditions (Table 3). The prevalence rates were higher in studies that included a greater proportion of persons over age 75 years (data not shown).

DISCUSSION

The prevalence of multimorbidity is greater than 60% worldwide and is probably greater than 80% among persons aged ≥85 years. The differences in prevalence rates between the studies were probably due to methodological differences rather than true differences; no 2 studies used the same methods, so there is no comparability. Key methodological issues included the type and number of chronic conditions included in the case definition of multimorbidity, how they were measured, the number of diseases defining multimorbidity, and the source population. The reported prevalence of multimorbidity was lower in studies that considered fewer than 10 chronic conditions. This sample of studies was not large enough to delineate other relations with methodological factors. Among older adults with any of the 15 index conditions, more than 90% had comorbid conditions from this set of conditions. Because the Medicare population under 65 years of age is eligible for the program largely because of disability, the multimorbidity prevalence rate in this age group is probably biased. The results for persons aged ≥65 years, however, are representative of the entire US population inasmuch as they come from a sample of more than 30 million adults enrolled in fee-for-service Medicare. The prevalence rates of individual chronic conditions, which are based on ascertainment from billing data, are generally consistent with other studies that have used such data. They tend to be higher than the rates reported in studies that used clinical methods; for example, the prevalence of heart failure in this study was 18% in the older adults, versus 5%–15% in national data from the United States (23).

The present review encompassed 17 population samples, including 5 that were more recent than a recent review by Fortin et al. (24). Fortin et al. incorporated only 13 population-based estimates, 2 of which did not meet our criteria. Although Fortin et al. rated articles for quality, all studies were rated as good, and apparently no studies were excluded on the basis of the quality assessment. Nevertheless, similar methodological concerns were observed, and similar conclusions were drawn.

Definition of multimorbidity

The underlying concept of a disease or health condition is that of a deviation from the normal state, with a dependency on basic science and convention to meet this criterion. For example, hypertension and hyperlipidemia are known risk factors for ischemic heart disease that were recognized as diseases at different times, and their definitions evolved differently. Chronicity is based on expected or actual duration of the condition. More recently, debate has arisen over whether to consider obesity a chronic condition, and, as a practical matter, its inclusion would considerably modify the prevalence of multimorbidity, especially in younger populations. Obesity was not included in the reviewed studies. The choice of how many chronic conditions to include, together with their individual prevalence rates, most strongly drives the prevalence of multimorbidity. Some authors recommend a minimum number, such as 12, but do not specify whether the choice should be based on prevalence or health burden (24). Although inclusion of a long list of less common conditions might increase the prevalence of multimorbidity, it would increase the complexity of the methods. Geriatric conditions, such as incontinence and falling, should be considered for inclusion in future multimorbidity research because they have routinely been omitted from the studies reviewed here.

Leave a Comment

(0 Comments)

Your email address will not be published. Required fields are marked *