Strengths and Limitations
This study made use of electronic primary care records which are broadly representative of the UK general population.25Availability of the information about prescribed medications in primary care records enabled us to select a large number of anonymized HRT users. The matched cohort study design and exclusion of the selected medical conditions from both cases and controls allowed us to estimate the effects of HRT on the survival of healthy users compared with healthy non-users. The use of multiple imputation techniques for missing records allowed us to include nearly all extracted patients in the analyses. Use of Weibull Double-Cox model enabled us to estimate the hazards of time-variant covariates. A wide range of available information in primary care records including comorbidities, treatment history, lifestyle factors and demographics allowed us to adjust for a high number of important confounders and the interaction between them. This study had a long-term average patient follow-up of almost 14 years.
The participants of this study received a wide variety of HRT preparations and doses, and thus these were not differentiated in the analyses. Although many important risk factors were adjusted for, there is likely to remain residual confounding by a number of other risk factors, such as age at menopause, parity, diet, and physical activity. These covariates were not adjusted for in the models as they were not reliably recorded in the health records. Duration of HRT use was not adjusted for as it may potentially introduce immortality bias (longer use is confounded with longer survival). The higher rates of diagnosed conditions in HRT users compared to non-users could be because the users visited the GP more frequently than the non-users as they were receiving the treatment, and hence their health status was checked and updated more often. Although THIN is broadly representative of the UK general population, due to high geographical clustering in THIN41, further research may be required to validate the results using data from other UK databases.