(150) Multimorbidity Ascertainment in Health Administrative Data and Role of The Time Window

Publication Number: 1195

Background: Patient multimorbidity is used to inform policy making and to study health outcomes in etiologic research. Multimorbidity measures based on health services administrative databases require the selection of an appropriate time-window (look-back window) for disease ascertainment.

Objectives: To assess the impact of various look-back windows (LWs) on multimorbidity prevalence and health outcomes prediction across multimorbidity measures, multimorbidity definitions and health outcomes.

Methods: In this population-based study, we assessed multimorbidity on April 1, 2019 for all residents >65 years in Quebec, Canada (n=1.4 million). We used linked hospitalization, physician and drug claims records from 1999 to 2020. We identified diseases with the International classification of diseases (ICD) system in hospitalization and physician claims for two lists of medical conditions: 1) the Comorbidity Combined index (CCI) (combination of Charlson and Elixhauser measures including 31 conditions associated with death); and 2) the Calderon multimorbidity measure (CMM) (60 conditions that regroup all chronic disease codes included in ICD). We considered LWs ranging from 1 to 20 years and evaluated their impact on 1) the prevalence of multimorbidity (defined as ≥2, ≥3 or ≥4 conditions); 2) the capacity (c-statistic) to predict 1-year outcomes: mortality, hospitalisation, polypharmacy, as well as general practitioner, specialist, or emergency department (ED) visits.

Results: The prevalence of multimorbidity increased continuously with increasing LWs, although it decelerated after 10 years (e.g., multimorbidity ≥2: LW=1y: [14%CCI; 20%CMM], LW=10y: [58%CCI; 84%CMM], LW=20y: [69%CCI; 92%CMM]). Impact of LWs on prediction capacity varied widely between lists of medical conditions, multimorbidity definitions, and the outcome considered. Maximal performance was reached with shorter LWs for the larger list of medical conditions (CMM) compared to the smaller list (CCI) (e.g., multimorbidity ≥4, ED visits: maximal c-statistic reached with LW=4y for CMM [0.714; 95%CI: 0.711-0.716] vs LW=10y for CCI [0.718; 95%CI: 0.716-0.720]). Performance was better with shorter LWs for multimorbidity ≥2 than multimorbidity ≥3 or ≥4 (e.g., hospitalisation, CCI: maximal c-statistic reached with LW=3y for multimorbidity ≥2 [0.677; 95%CI: 0.676-0.679] vs LW=11y for multimorbidity ≥4 [0.681; 95%CI: 0.679-0.682]).

Conclusions: Inappropriate LWs may reduce capacity to interpret multimorbidity indicators and to control for confounding bias. LWs impact both multimorbidity prevalence and outcome predictions, and is widely influenced by the multimorbidity measure, multimorbidity definition and health outcome studied.