A substantial proportion of CHD events in primary care are recorded using non-specific codes, which may artificially affect disease incidence estimates and highlights the need for careful code selection in EHR-based research.
OBJECTIVES: To evaluate the coding, recording and incidence of coronary heart disease (CHD) in primary care electronic medical records. METHODS: Data were drawn from the UK General Practice Research Database. Analyses evaluated the occurrence of 271 READ medical diagnostic codes, including categories for 'Angina', 'Myocardial Infarction', 'Coronary Artery Bypass Grafting' (CABG), 'percutaneous transluminal coronary angioplasty' (PCTA) and 'Other Coronary Heart Disease'. Time-to-event analyses were implemented to evaluate occurrences of different groups of codes after the index date. RESULTS: Among 300,020 participants aged greater than 30 years there were 75,197 unique occurrences of coronary heart disease codes in 24,244 participants, with 12,495 codes for incident events and 62,702 for prevalent events. Among incident event codes, 3,607 (28.87%) were for angina, 3,262 (26.11%) were for MI, 514 (4.11%) for PCTA, 161 (1.29%) for CABG and 4,951 (39.62%) were for 'Other CHD'. Among prevalent codes, 20,254 (32.30%) were for angina, 3,644 (5.81%) for MI, 34,542 (55.09%) for 'Other CHD' and 4,262 (6.80%) for CABG or PCTA. Among 3,685 participants initially diagnosed exclusively with 'Other CHD' codes, 17.1% were recorded with angina within 5 years, 5.6% with myocardial infarction, 6.3% with CABG and 8.6% with PCTA. From 2000 to 2010, the overall incidence of CHD declined, as did the incidence of angina, but the incidence of MI did not change. The frequency of CABG declined, while PCTA increased. CONCLUSION: In primary care electronic records, a substantial proportion of coronary heart disease events are recorded with codes that do not distinguish between different clinical presentations of CHD. The results draw attention to the need to improve coding practice in primary care. The results also draw attention to the importance of code selection in research studies and the need for sensitivity analyses using different sets of codes.
Bhattarai et al. (Thu,) studied this question.