Validation of an Algorithm Identifying Incident Primary Immune Thrombocytopenia in the French National Health Insurance Database

Introduction: Population-based studies using health insurance databases are very useful for epidemiological studies, notably for rare diseases such as immune thrombocytopenia (ITP). French health insurance databases, called SNIIRAM (Système d'Information Inter-Régimes de l'Assurance Maladie) cover the entire French population (66 million inhabitants).We developed an algorithm that identifies incident primary ITP cases in the SNIIRAM. The aim of this study was to evaluate the positive predictive value (PPV) of this algorithm.

Methods:The SNIIRAM is an electronic database prospectively recording demographic, hospitalization, long term disabling disease (LTD) and out-hospital drugs dispensing as well as other healthcare data.

The algorithm used to identify newly diagnosed primary ITP patients follows several steps: 1) extraction of patients with at least one ITP code (D69.3 code of the International Classification of Disease, version 10 -ICD-10) as primary diagnosis during a hospital stay or as LTD; 2) exclusion of patients with ambiguous or contradictory codes, i.e.other D69 ICD-10 codes as LTD or hospital diagnosis between 12 months before and 6 months after the first ITP code, except D69.6 (thrombocytopenia, unspecified) and D69.9 (hemorrhagic condition, unspecified); 3) refining of the presumed date of diagnosis searching ITP drug dispensing in the six-month period before the first ITP code. A six-month wash out period is used to define incident ITP; 4) lastly, secondary ITPs are defined by a LTD or a hospitalization with codes corresponding to hematological malignancies, chronic viral infections and connective tissue diseases between 12 months before and 6 months after the presumed date of diagnosis.

For the present validation study, we applied this algorithm to regional SNIIRAM data of the Midi-Pyrénées region (South of France, 3 million inhabitants) to identify incident ITP patients between July 1, 2012 and June 30, 2014. Medical charts were independently reviewed by two investigators.

We calculated the PPVs of: 1) the diagnosis of ITP; 2) the fact that ITP was incident or not (compared with the true date of diagnosis and the true date of first ITP symptoms); 3) the identification of secondary ITPs.

Results: We identified 200 incident ITP patients in the Midi-Pyrénées regional SNIIRAM during the study period. We could assess by medical chart review 168 cases. At ITP diagnosis, 70.4% of the patients were adults (≥18 years). Median age was 2.8 years in children and 63.0 years in adults. The male:female sex-ratio was 0.83. Overall, 66.6% of the patients had bleeding symptoms. The median platelet count was 10 x 10⁹/L.

After medical chart review, 161 cases were true ITP yielding a PPV of 95.8% (95% confidence interval - 95% CI: 92.8-98.8). Among them, 128 were incident according to the date of symptom onset and 134 were incident according to the true diagnosis date yielding PPVs of 79.5% (95% CI: 73.2-85.7) and 83.2% (95% CI: 77.4-89.0), respectively. As expected, the PPV for newly diagnosed ITP was significantly lower in case of low platelet count as first symptom compared tobleeding revealing ITP (respectively, 65.4%, 95% CI: 52.5-78.3 and 92%, 95% CI 86.9-97.0, p=0.08). The median time from the date of first symptoms and the estimated date of ITP onset by the algorithm was 3 days (interquartile range - IQR: 0 to 20). The median time from the true date of diagnosis and the estimated date of ITP diagnosis by the algorithm was 0 days (IQR: 0 to 15).

Ten patients were identified as having a secondary ITP using the algorithm: 4 lymphoid neoplasms (2 B-cell lymphoma, 1 Hodgkin lymphoma and 1 chronic lymphoid leukemia), 4 connective tissue diseases (sarcoidosis, rheumatoid arthritis, systemic sclerosis and Sjögren syndrome, 1 case each) and 2 chronic infections (HIV and HCV). There was no misclassification after medical chart review. On the contrary, 7 secondary ITPs identified by medical chart review were not detected by the algorithm: 3 CMV and 1 EBV infections, 1 Waldenström disease, 1 plasmocytoma and 1 myelodysplastic syndrome.

Conclusions: This algorithm showed good PPV in identifying incident primary ITP patients and could be extended to other similar databases. The estimated date of ITP onset may be delayed in case of isolated chronic thrombocytopenia because platelet count results are not recorded in the database. The SNIIRAM is not a useful database to detect virus-associated secondary ITP.