Original ArticleSpecific comorbidity risk adjustment was a better predictor of 5-year acute myocardial infarction mortality than general methods
Introduction
Cardiovascular disease (CVD) is the single leading cause of death in the province of British Columbia, Canada, accounting for >35% of deaths [1]. Because the burden of illness due to CVD affects a great number of individuals, assessment of the outcomes of acute myocardial infarction (AMI) is extremely important. Population-based administrative databases are useful for this purpose because they cover the entire population, giving significant weight to the generalizability and applicability of any research findings. Moreover, administrative data are an efficient data source, compared to primary data collection, for clearly defined clinical conditions in the context of long periods of follow-up.
Severity of illness is one important variable that researchers must measure when studying the outcomes of AMI patients. There are limitations in using administrative data to assess severity of illness because important indicators of severity such as electrocardiographic changes, elevation of creatinine phosphokinase (CPK), and other physiological indices are not available. Nonetheless, although short-term mortality is most likely to be related to the physiological severity of AMI, long-term mortality, which is the primary outcome of this study, is more likely related to concurrent or underlying comorbidity [2], which is readily identified in administrative data.
Our objective was to determine how best to adjust for comorbidities when investigating post-AMI mortality using administrative data. Results from this analysis can assist clinicians in assessing patient prognosis, by determining which comorbidities and methods of measuring them are most predictive of mortality post AMI.
To identify the range of methods available, we consulted the literature on risk-adjustment methods. A few researchers have developed indices for risk adjustment in AMI patients [2] and AMI report cards [3], [4] using hospital discharge databases [5]; however, these models are limited for several reasons. A model developed by Normand et al. [2], a 40-variable prediction rule using U.S. Medicare data, is large and complicated, which renders it unappealing and impracticable to health services researchers. Methods developed in California and Pennsylvania [3], [4] used variables with American-specific response options such as race (black vs. white). This approach is not relevant for studies in British Columbia. More importantly, the prediction rules developed by these researchers focused only on predicting in-hospital and 30-day mortality [3], [4], and not longer-term mortality, which was the outcome of interest in this present study. They also required separate models for direct admission or transferred-in patients [4] and separate models for individuals with no prior hospital admission and those previously hospitalized, which further complicated the analyses [3]. In contrast to these complicated models, the simple Ontario AMI prediction rule (OAMIPR) developed in Ontario by Tu et al. [6] is more applicable to the British Columbia population. Thus, the comorbidities that were found in the OAMIPR were chosen as one method of risk adjustment for this study. This validated model predicts 30-day and 1-year mortality with nine comorbidities, in addition to age and sex: shock, diabetes, congestive heart failure, cancer, cerebrovascular disease, pulmonary edema, acute renal failure, chronic renal failure, and cardiac dysrhythmia.
We also assessed the Charlson comorbidity index, a common method of risk adjustment—albeit not specific for individuals with AMI. It is based on relative risks of mortality for 19 conditions observed during a longitudinal study of 559 internal medicine patients [7]. In the development of the index, any disease generating a relative risk of >1.2 and <1.5 was retained and weighted as 1; a weight of 2 was given for a relative risk of 1.5 to ≤2.5; a weight of 3 for relative risk of 2.5 to ≤3.5; and a weight of 6 for two conditions with a relative risk of ≥6. The sum of the weights for each individual is calculated. We used the D'Hoore adaptation of the Charlson Index [8], because the D'Hoore adaptation uses three-digit ICD-9 codes, and the data analyzed in this study have been shown to have higher validity when only the first three digits are used (W. Hu, unpublished report, July 1996). The comorbidities, their ICD-9 codes, and corresponding weights used in the D'Hoore adaptation are shown in Table 1.
A third method for controlling for comorbidities is a measure of the number of distinct comorbidities present. Schneeweiss et al. [9] used administrative data from British Columbia for a population of individuals 65 years of age or older who had filled at least one prescription for an angiotensin-converting enzyme inhibitor or calcium channel blocker. Schneeweiss et al. [9] compared this method to other common methods of controlling for comorbidities (including the Charlson Index and other scores based on it, as well as scores based on outpatient drug utilization data). They found that the number of distinct comorbidities performed just as well as other risk-adjustment methods in predicting 1-year mortality. This method is unique in that (i) it does not give greater weight to illnesses that have a stronger impact on survival, as the Charlson Index does; (ii) it is very general; and (iii) it was not developed specifically for predicting mortality following AMI (as is, for example, the list of nine comorbidities identified by Tu et al. [6]). Because of its relative simplicity, we thought it important to determine how well this approach performs compared to the other methods, including the OAMIPR [6] and the Charlson Index [7].
In summary, the present study compared three approaches to risk adjustment to predict mortality post AMI: the OAMIPR [6], the Charlson Index [7], and the number of distinct comorbidities [9]. The goal was to determine the best method, rather than to generate the ultimate predictive model of mortality post AMI.
Section snippets
Data sources
The British Columbia Linked Health Database (BCLHD) was used to identify the study population and to construct the variables. The BCLHD is a population-based data resource developed and maintained by the Centre for Health Services and Policy Research (CHSPR). CHSPR acts as the custodian of and access point for the various data holdings of the BCLHD, which remain under the stewardship of the agency that originally collected them. CHSPR prepares data for analysis for approved research projects.
Results
All but 9 of the 135 models had nonsignificant Hosmer–Lemeshow goodness-of fit statistics. We were not particularly concerned about those models that were significant, because the correlates included in the model were determined a priori and especially because there was not an obvious pattern associated with the very few models that demonstrated poor overall model fit.
The C-statistic results for each of the five cohorts are shown in Table 4. Model 3, which includes the OAMIPR comorbidities [6]
Discussion
We compared three approaches to risk adjustment that can be used as methods of controlling for confounding in epidemiological studies using administrative data. The OAMIPR [6], a disease-specific risk-adjustment method, had the highest C-statistic and R2 values, compared to more general risk-adjustment methods, such as D'Hoore's adaptation of the Charlson Index [8] or the number of distinct comorbidity diagnoses. The OAMIPR comorbidities [6] measured during a longer period after the index AMI
References (18)
- et al.
Development and validation of a claims based index for adjusting for risk of mortality: the case of acute myocardial infarction
J Clin Epidemiol
(1995) - et al.
Development and validation of the Ontario acute myocardial infarction mortality prediction rules
J Am Coll Cardiol
(2001) - et al.
A new method of classifying prognostic comorbidity in longitudinal studies: development and validation
J Chronic Dis
(1987) - et al.
Practical considerations on the use of the Charlson Comorbidity Index with administrative data bases
J Clin Epidemiol
(1996) - et al.
The evolving clinical status of patients after a myocardial infarction: the importance of post-hospital data for mortality prediction
J Clin Epidemiol
(1996) - British Columbia Ministry of Health and Ministry Responsible for Seniors, and Heart and Stroke Foundation of BC &...
- et al.
Report on heart attack 1991–1993. Vol. 2. Technical guide
(1997) Focus on heart attack in Pennsylvania: research methods and results
(1996)- et al.
Institute for Clinical Evaluative Sciences in Ontario. Cardiovascular health and services in Ontario: an ICES atlas
(1999)
Cited by (44)
Design and outcomes of the Patient Centered Outcomes Research Institute coronary heart disease cohort study
2018, Contemporary Clinical Trials CommunicationsDistinct trajectories of multimorbidity in primary care were identified using latent class growth analysis
2014, Journal of Clinical EpidemiologyCitation Excerpt :Studies have generally measured multimorbidity at a single point or period in time, and thus ignored changes in multimorbidity over time. Two prospective studies found that changes in comorbidity significantly increased long-term mortality risks in people with breast cancer [38] and heart attack [39]. A Dutch study found that developing multimorbidity over time was associated with poorer physical functioning, but there was a little association with psychological health status [40].
Review of methods used to measure comorbidity in cancer populations: No gold standard exists
2012, Journal of Clinical EpidemiologyCitation Excerpt :For example, different conditions are likely to be important in terms of their impact on outcomes for patients with breast cancer compared with those with congestive heart failure. For this reason, a number of authors have suggested that disease-specific indices, such as cancer-specific measures are preferable to general ones [14–16]. However, whereas generic comorbidity scales may lack specificity, individualized comorbidity indices lack comparability across research settings.
Impact of Socioeconomic Deprivation and Area of Residence on Access to Coronary Revascularization and Mortality After a First Acute Myocardial Infarction in Québec
2012, Canadian Journal of CardiologyCitation Excerpt :For outcomes, we assessed 6-month coronary angiography rates, revascularization rates, 30-day, and 1-year mortality rates, all following the index AMI, considered as the time of admission. For 30-day mortality, we compared predicted22–24 vs observed rates in patients by level of deprivation. Trends for baseline discrete data and for 1-year mortality were compared using Mantel-Haenszel χ2 and continuous data using analysis of variance.
Impact of comorbidities on in-hospital mortality from acute myocardial infarction, 2003-2009
2011, Revista Espanola de Cardiologia