BACKGROUND: The validity of algorithms to identify hospitalizations for heart failure (HF) among patients with chronic obstructive pulmonary disease (COPD) in the Clinical Practice Research Datalink (CPRD) in the United Kingdom has not been described.
OBJECTIVES: To validate potential cases (including deaths) of hospitalization for HF in patients with COPD in the CPRD.
METHODS: Hospitalizations for HF in a cohort of new users of selected COPD medications (September 2012-June 2017) were identified electronically through: (1) primary and secondary discharge diagnosis for HF in the Hospital Episode Statistics (HES), (2) HF recorded as cause of death in the Office for National Statistics (ONS) and a code for hospitalization within 30 days before in the CPRD General Practitioner Online Database (GOLD), and (3) HF diagnosis and a record for hospitalization within 30 days in the CPRD-GOLD when linkage to HES/ONS was not available. A questionnaire (Qx) was sent to the general practitioners (GPs) to confirm diagnosis. For cases with conflicting information from Qx, adjudication was performed through manual clinical review of the patient profiles. The positive predictive value (PPV) was calculated as an indicator of the validity of HF diagnosis.
RESULTS: There were 2,283 potential HF cases identified in 51,319 individuals with COPD aged ≥ 40 years. A Qx was sent to the GPs for 1,176 potential HF cases identified from active practices. The response rate was 69.7%. Among the Qx received, 656 Qx (55.8%) included evaluable information. Of these 656 potential cases, 434 were confirmed (PPV = 66.2%). The cases confirmed were 97 of 102 cases (PPV = 95.1%) identified through HES primary discharge diagnosis, 168 of 350 cases (PPV = 48.0%) identified through HES secondary discharge diagnoses, and 169 of 204 cases (PPV = 82.8%) identified through CPRD GOLD diagnosis code and a hospitalization code. PPVs were similar among patients with or without prior history of hospitalization for HF.
CONCLUSIONS: Among patients with COPD, the algorithms used to identify hospitalizations for HF through HES primary discharge diagnosis and through GOLD HF diagnosis and hospitalization codes had a high PPV. The algorithm that identified cases through secondary discharge diagnosis had a lower PPV but contributed to a high proportion of the total cases confirmed.
