Background
Administrative data are commonly used for surveillance, assessment of health care utilisation and evaluation of health outcomes for people with dementia. However, the accuracy of ICD-10 coded hospital and mortality data in identifying mild cognitive impairment (MCI) and dementia is largely unknown.
Aim
To calculate the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of MCI and dementia coding.
Methods
Participants of the population-based Sydney Memory and Ageing (MAS) study underwent 2-yearly neuropsychological testing. Cognitive status was classified into normal, MCI and dementia at each assessment by a consensus panel of clinicians providing a robust diagnosis. MAS records were probabilistically linked to NSW hospitalisation and death records for 2001-2014. MCI (ICD-10: F06.7, G31.84) and dementia (ICD-10: F00, F01, F02, F03, F05.1, G30, G31.1) were identified from the administrative data.
Results
There were 14,467 hospitalisations for the 1,026 participants and 273 deaths. Over the study period 553 (53.9%) participants had a diagnosis of MCI and 105 (10.2%) of dementia. MCI was not recorded on any hospital or death record. For dementia, sensitivity was low (29.1%) but specificity high (99.4%) for hospital records. PPV was 59.4% and NPV 97.9%. Only 5 (6.2%) of participants with dementia, had dementia recorded on every hospitalisation following their diagnosis. Similarly, sensitivity was very low (17.0%) and specificity high (96.5%) for death records. PPV was 50.0% and NPV 84.8%.
Conclusion
MCI is not recorded and dementia is under-recorded in administrative data. To maximise identification of dementia an ‘ever-identified’ approach is recommended.