Single index of multimorbidity did not predict multiple outcomes

Abstract

Background and Objectives

Measurement of multimorbidity and comorbidity is important in epidemiologic and health services research. The aim of this research was to derive a generic multimorbidity index based on patient self-report, incorporating severity, for predicting a range of outcomes.

Methods

The dataset was obtained from a trial including 1,541 Veterans and war widows aged 70 years and over. The survey included sociodemographics, hospital admissions, SF-36, and information on deaths was obtained. The methods of Charlson were used to derive Multimorbidity Indices.

Results

All indices predicted quality of life, with decreasing quality of life for each increase in multimorbidity category. Multimorbidity scores incorporating severity significantly contributed to the prediction of mortality, hospital admission, and follow-up quality of life, regardless of adjustment for baseline quality of life.

Conclusions

Our results indicate that a single index cannot predict a variety of relevant outcomes. Consequently, research undertaken to assess the impact of intervention or illness on health outcomes should use an index that is valid for predicting the specific outcome of interest.

Introduction

The measurement of multimorbidity (the coexistence of multiple diseases in the same individual [1]) and comorbidity (the coexistence of chronic conditions within the context of an index condition [2], [3]) is important in epidemiologic and health services research because coexisting morbidities have been shown to be associated with many important health outcomes such as quality of life (QoL) [4], activities of daily living [5], health service utilization, and mortality [6], [7], [8], as summarized in a review of the causes and consequences of coexisting morbidities by Gijsen et al. [8]. Measurement of multiple coexisting conditions is particularly important in observational studies where it is necessary to adjust for confounding factors such as disease states and severity.

Numerous methods for measurement of multimorbidity have been reported [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], with various terms (including comorbidity, multiple morbidity, and multimorbidity) being used interchangeably by some authors [1], [3], [9]. Perhaps the best known index of this type is the Charlson Comordibity Index [10]. This instrument was developed from an initial list of all possible comorbid conditions identified from hospital medical records. From this list a catalogue of comorbidities was developed. Charlson et al. [10] then used the Cox Proportional Hazards Regression [18] to obtain a set of weightings for each condition in terms of their ability to predict mortality. These weightings could be summed to provide a comorbidity score. The weighted index was validated in a cohort of women with breast cancer and found to be predictive of mortality [11]. Since the development of the Charlson Index further work has been undertaken to adapt the index to other research settings, including use with administrative data sets [12], [13] and use as a self report measure [14]. The Katz [14] instrument was one of the first measures to be adapted for use as a self-administered instrument.

It has been recently noted that there are several advantages to gathering information on morbid conditions from patients themselves rather than from medical records [15]. First, health administrative and medical record data rely on the information being recorded in the medical record and coded in the administrative data set. Second, often researchers cannot obtain access to medical records or administrative data. Third, administrative-based instruments may only be appropriate for hospitalized patients, and may not be appropriate for population research. Although the Katz [14] instrument is practical to administer outside a health care setting, it has been criticized because it does not include all potentially relevant conditions, and it does not assess severity [15]. Some more recent but less well-known self-reported instruments have been developed to incorporate current severity of conditions [15]. For example, Crabtree et al. [15] developed the CmSS, a simple interviewer administered tool for use in older people as an objective measure of the presence of comorbid disease, which includes the patient's perception of the severity of the associated symptoms.

A further limitation of many of the validated indices previously reported is that the scoring systems were developed for a single outcome (usually mortality) in a specific patient group, and the generalizability of these indices to other populations in other settings and to predict other outcomes has not been demonstrated [16], [17]. Other than mortality, relevant outcomes for epidemiologic and health services research can include hospital admission and QoL. Both these outcomes are significant in terms of health care cost and burden of illness. Several studies have shown that QoL and multimorbidity are highly correlated [4], [19], [20]. Although other indicators of morbidity, such as visits to the family doctor or other health care provider, have previously been used [8], these indicators represent a lower impact than mortality, hospital admissions, and QoL as indicators, as for instance, doctor visits may be preventative or routine rather than always reflecting health outcomes.

Our aim in this research was to attempt to derive a generic multimorbidity index (i.e., without referral to an index condition [1]) based on patient self-report and incorporating self-reported severity, which can be used for predicting a range of outcomes including mortality, hospital admission, and health-related QoL. We were also interested in assessing whether a multimorbidity index added to the prediction of these outcomes beyond the predictive ability of baseline QoL scores. We expected that given the previously observed high correlation between QoL and multimorbidity [4], [8], adjustment for QoL may partially adjust for multimorbidity. Thus, we sought to answer the following research questions:

• What is the association between the different morbidities and death, hospital admission, and QoL, as separate outcomes?

• Can a simple index of multimorbidity, which incorporates the measure of association between each morbidity and death, predict hospital admission?

• Can a simple index of multimorbidity, which incorporates the measure of association between each morbidity and hospital admission, predict death?

• Does addition of a severity measure to the multimorbidity index improve prediction of death or of hospital admission?

• What impact does adjustment for baseline QoL have on the predictive models?