(11) The Completeness of The Swedish National Patient Register in Terms of Cancer Diagnoses

Publication Number: 685

Background: The Swedish National Health Registers are used for a variety of purposes, including pharmacoepidemiological research due to their large and linkable data. The National Patient Register (NPR) covers information on diseases and medical procedures for the entire Swedish population. The use of NPR to identify cancer risk in safety studies is desired, but NPR has not been validated to date. High completeness of the Swedish Cancer Register (SCR) has been reported, however, its usefulness is limited by its long production time and annual updates. If the consistency of NPR for reporting cancer diagnoses compared to SCR can be shown, it might be widely used in research that requires rapid reporting.

Objectives: The overall aim of this study was to evaluate the consistency of NPR for cancer diagnoses as compared to SCR.

Methods: This was a cross-sectional study with the entire Swedish population as the source population. The study population was patients with at least one cancer registered in SCR between 1 January 2018 and 30 September 2020. The comparison was made between cancer diagnoses identified by the International Classification of Disease 10th version and by the International Classification of Disease for Oncology 3rd version reported in NPR and SCR, respectively. If the same cancer recorded in SCR was found in NPR within +/- 90 days it was considered consistent. Cancer was categorized into 13 classes according to the location of the tumor, excluding lymphoid, hematopoietic, and multiple sites. The frequency of cancers reported in NPR and SCR were compared overall and by category, using SCR as the gold standard. Accuracy measurements including positive predictive value (PPV), sensitivity, the area under the receiver operating characteristic curve (AUC), false positive rate (FPR), and false negative rate (FNR) were calculated.

Results: In total 223,843 cancer diagnoses in SCR and 124,466 in NPR were identified. The 4 most common cancer types in the registers were skin, digestive organs, male genital organs, and breast. The higher accuracy measurements were observed for breast, male genital organs, and oral cancers (PPV: 87.5-97.4%, sensitivity: 82.2-91.0%, AUC: 0.91-0.95, FPR: 0.18-1.5, FNR: 9.0-17.8). The measurements were lower for skin, female genital, and ill-defined/secondary cancers (PPV: 8.8-47.8%, sensitivity: 19.9-32.3, AUC: 0.59-0.66, FPR: 0.39-6.6, FNR: 69.1-80.1).

Conclusions: Consistency of cancer diagnoses in NPR depends on cancer type. The observed high consistency may be explained by the availability of screening. A potential explanation for the lower consistency may be differences in the coding and reporting processes. The mechanisms behind this difference in consistency remain unclear and further investigation is needed.