Abstract
Background: A large clinical dataset of vitamin B12 measurements (over 12,000 patients) and Schilling tests (over 700 patients) has been assembled from participating Manitoba laboratories (1995–2000) in order to assess clinical outcomes (as defined from administrative health data) in relation to tests of cobalamin [Cbl] deficiency. In order to control for Cbl replacement, it is necessary to confirm that Cbl replacement can be accurately determined from administrative data sources such as the Drug Program Information Network (DPIN) and/or medical claims for therapeutic injections. DPIN is known to be reliable for oral prescription drugs, but parenteral Cbl may be more difficult to identify and non-prescription oral Cbl would not be recorded. We believe that the preferred route of Cbl administration in Manitoba was parenteral during 1995–2000, but this needs to be confirmed.
Methods: Cases (expected to have a high likelihood for Cbl replacement) consisted of all adult hospital discharges between 1995 and 2000 with a diagnosis of pernicious anemia or other Cbl deficiency anemia (ICD-9-CM 281.0, 281.1) who survived for at least one year post-discharge without personal care home admission (n=388). Matched controls (3 controls for each case) with a low likelihood for Cbl replacement were similarly defined from hospital discharges without any ICD-9-CM neurologic or haematologic diagnosis (n=1,164). Controls were matched with cases for gender, birth year and hospitalization year. For each case and control the number of Cbl dispensations (ATC code B03BA01) and therapeutic injections (tariff code #8954) were tabulated during the first three years post-discharge. Sensitivities (fraction of cases receiving Cbl replacement) and specificities (fraction of controls not receiving Cbl replacement) were calculated for different cutoffs of Cbl dispensations, therapeutic injections and combinations according to year of diagnosis (YOD) and overall. Weighted linear regression between YOD and sensitivity was performed to look for an effect of recent oral Cbl replacement.
Results: Sensitivity for the a priori operational definition (≥2 Cbl dispensations) was 0.59±0.02 with specificity 0.96±0.01. Using one or more Cbl dispensations improved sensitivity to 0.71±0.02 with similar specificity of 0.95±0.01. Therapeutic injections alone were less useful (≥1 injection sensitivity of 0.65±0.02 with specificity of 0.92±0.01, ≥2 injections sensitivity of 0.61±0.02 with specificity of 0.95±0.01). The combination that maximized sensitivity + specificity was ≥1 dispensation or ≥2 therapeutic injections (sensitivity 0.80±0.02, specificity 0.93±0.01). Sensitivity measurements did not show any significant relationship with YOD.
Conclusions: The optimal definition for Cbl therapy based upon administrative health data appears to be at least one Cbl dispensation identified in the DPIN database or at least two therapeutic injections identified from medical claims. In assessing clinical outcomes in relation to Cbl testing and replacement, we propose to validate findings derived with the primary combined definition through secondary analyses using independent definitions of Cbl replacement (≥1 dispensation alone or ≥2 therapeutic injections alone). Overall sensitivity was lower than expected but may reflect that some cases are not prescribed Cbl replacement post-discharge. Sensitivity was unrelated to YOD arguing against oral Cbl replacement as the explanation.
Author notes
Corresponding author
