Validation of algorithms identifying diagnosed obstructive sleep apnoea and narcolepsy in coded primary care and linked hospital activity data in England

Purpose

To assist sleep epidemiology research, we created and tested the accuracy of five algorithms identifying diagnosed Obstructive Sleep Apnoea (OSA) and narcolepsy in routinely collected data from England (01/01/1998–29/03/2021).

Methods

The primary algorithm identified the first coded record in Clinical Practice Research Datalink (CPRD) primary care or linked hospital admissions data as an incident diagnosis of OSA (n = 92,222) or narcolepsy (n = 1072). Alternative algorithms required codes in CPRD, both datasets, or an additional proximate possible-sleep-related outpatient visit or excessive daytime sleepiness drug prescription (narcolepsy only). Staff in 73/1574 CPRD practices completed online questionnaires for a convenience sample of 144 OSA and 101 narcolepsy cases. We estimated Positive Predictive Values (PPVs) describing the proportion of cases confirmed by a gold standard hospital specialist diagnosis, the percentage of gold standard cases from the primary algorithm retained with alternative algorithms, and time between specialist and recorded diagnosis dates.

Results

Using the primary algorithm, the PPV (95 % CI) was 75.3 % (69.2–81.3) and 65.2 % (57.0–73.4) for OSA and narcolepsy, respectively: 80.6 % and 62.7 % of confirmed cases were recorded within 6 months of the specialist diagnosis. The CPRD-only algorithm increased the PPV to 85.3 (77.3–91.4, OSA) and 71.0 (58.8–81.3, narcolepsy) and retained high proportions of gold standard cases. Requiring additional outpatient or prescribing data increased PPVs, and for OSA improved diagnostic date accuracy, but omitted a high proportion of gold standard cases.

Conclusion

Highly accurate OSA diagnoses can be identified in routinely collected data. Recorded cases of narcolepsy are moderately accurate, but diagnosis dates are not.