Identification of diabetes, heart disease, hypertension and stroke in mid- and older-aged women: Comparing self-report and administrative hospital data records.

AIM: To estimate the prevalence of diabetes, heart disease, hypertension and stroke in self-report and hospital data in two cohorts of women; measure sensitivity and agreement between data sources; and compare between cohorts. METHODS: Women born between 1946-1951 and 1921-1926 who participated in the Australian Longitudinal Study on Women's Health (ALSWH); were New South Wales residents; and admitted to hospital (2004-2008) were included in the present study. The prevalence of diabetes, heart disease, hypertension and stroke was estimated using self-report (case 1 at latest survey, case 2 across multiple surveys) and hospital records. Agreement (kappa) and sensitivity (%) were calculated. Logistic regression measured the association between patient characteristics and agreement. RESULTS: Hypertension had the highest prevalence and estimates were higher for older women: 32.5% case 1, 45.4% case 2, 12.8% in hospital data (1946-1951 cohort); 57.8% case 1, 73.2% case 2, 38.2% in hospital data (1921-1926 cohort). Agreement was substantial for diabetes: κ = 0.75 case 1, κ = 0.70 case 2 (1946-1951 cohort); κ = 0.77 case 1, κ = 0.80 case 2 (1921-1926 cohort), and lower for other conditions. The 1946-1951 cohort had 2.08 times the odds of agreement for hypertension (95% CI 1.56 to 2.78; P < 0.0001), and 6.25 times the odds of agreement for heart disease (95% CI 4.35 to 10.0; P < 0.0001), compared with the 1921-1926 cohort. CONCLUSION: Substantial agreement was found for diabetes, indicating accuracy of ascertainment using self-report or hospital data. Self-report data appears to be less accurate for heart disease and stroke. Hypertension was underestimated in hospital data. These findings have implications for epidemiological studies relying on self-report or administrative data.