The Agreement Between Diagnoses as Stated by Patients and Those Contained in Routine Health Insurance Data—Results of a Data Linkage Study

Abstract

Background: The frequency of medical diagnoses is a figure of central importance in epidemiology and health services research. Prevalence estimates vary depending on the underlying data. For a better understanding of such discrepancies, we compared patients' diagnoses as reported by themselves in response to our questioning with their diagnoses as stated in the routine data of their health insurance carrier.

Methods: For 6558 adults insured by BARMER, one of the statutory health insurance carriers in Germany, we compared the diagnoses of various illnesses over a twelve-month period, as reported by the patients themselves in response to our questioning (October to December 2021), with their ICD-10-based diagnosis codes (Q4/2020-Q3/2021). The degree of agreement was assessed with two kappa values, sensitivity, and specificity.

Results: The patients' stated diagnoses of diabetes and hypertension agreed well or very well with their diagnosis codes, with kappa and PABAK values near 0.8, as well as very high sensitivity and specificity. Moderately good agreement with respect to kappa was seen for the diagnoses of heart failure (0.4), obesity, anxiety disorder, depression, and coronary heart disease (0.5 each). The poorest agreement (kappa ≤ 0.3) was seen for posttraumatic stress disorder, alcohol-related disorder, and mental and somatoform disorder. Agreement was worse with increasing age.

Conclusion: Diagnoses as stated by patients often differ from those found in routine health insurance data. Discrepancies that can be considered negligible were found for only two of the 11 diseases that we studied. Our investigation confirms that these two sources of data yield different estimates of prevalence. Age is a key factor; further reasons for the discrepancies should be investigated, and avoidable causes should be addressed.

Figure

Proportion of medical diagnoses as stated by patients in the survey and/or contained in routine data as well as the agreement in diagnoses between the two data sources using Cohen’s κ, PABAκ, sensitivity, and specificity. The data relate to the 12 months preceding the time of the survey or to the 4th quarter of 2020–3rd quarter of 2021 (see data table in eTable 1 for illustration) The proportion of individuals identified in at least one of the two data sources is calculated as the sum of the three proportions listed. Sensitivity corresponds to the percentage of people with a documented diagnosis in the routine data who also state the respective diagnosis in the survey, in relation to all people who have the diagnosis documented in the routine data. Specificity is the percentage of people without a documented diagnosis in the routine data who also do not state a respective diagnosis in the survey, in relation to all people who do not have a corresponding diagnosis documented in the routine data. κ, Cohen’s kappa measure of agreement; PABAκ, prevalence- and and bias-adjusted kappa; PTSD, post-traumatic stress disorder