(152) Evaluating Methods to Define Place of Residence in Administrative Data and the Impact on Observed Associations

Background: Literature supports the association between where an individual resides and access to healthcare services, drug therapy utilization, and health outcomes. However, evidence is lacking on the methods used to define residence and how this may impact measures of association.

Objectives: Quantify the impact of different methods to define residence on the association with all-cause mortality.

Methods: We identified a cohort of adult new metformin users using administrative health data from Alberta, Canada between 2008 and 2019. An individual’s place of residence (rural/urban/metropolitan) was identified as the exposure using postal codes at different time points. Time points were defined using 3 different approaches including time-fixed (at 1-year and 3-year lookbacks from first metformin), time-varying (during the outcome observation window), and nested case control (residence closest to the index date after identifying cases and controls). Cox proportional hazard models were used to conduct analyses for the time-fixed and time-varying approaches; a multivariable logistic regression model was used to conduct the analysis for the nested case control approach. All models included the same covariates (age, sex, measures of healthcare utilization, and diabetes complications severity index score) and used metropolitan as the comparator.

Results: A total of 157,146 new metformin users were identified. The mean age was 55 years, 57% were male, and 8,444 (5%) deaths occurred during the mean follow up of 4.7 years. Residence was distributed as 26% rural, 10% urban, and 64% metropolitan for all exposure definitions. There were few instances of moving after first metformin; at the first instance of moving, 2.6% of individuals moved to a smaller centre (metropolitan to urban or rural; or urban to rural) and 3.1% of individuals moved to a larger centre (rural to urban or metropolitan; or urban to metropolitan).

Rural residence was associated with increased mortality in all models. For the time-fixed approaches, risk of death for rural residents was aHR:1.18; 95% CI:1.12-1.24 and aHR:1.17; 95% CI:1.10-1.24 when residence was defined using 1-year and 3-year lookbacks, respectively. The time-varying approach showed a risk of death of aHR:1.17; 95% CI:1.12-1.23 for rural residents. The nested case control approach resulted in an odds of death of aOR:1.17; 95% CI:1.12-1.23 for rural residents.

Conclusions: The method used to define residence has minimal impact on measures of association, possibly owing to the infrequency with which people move within Alberta. Further studies are required to discern causality between residence and mortality.