An automated algorithm for consolidating dates of diagnosis from multiple sources.

Background: Multiple dates of diagnosis are often received from different reporting sources at a central cancer registry. Resolving these inconsistencies can be a labor-intensive task. To our knowledge, no algorithms for the consolidation of diagnosis dates have been published. We present such an algorithm here.

Methods: The algorithm uses a "take the best" heuristic approach, incorporating the reported dates of diagnosis, class of case, service type (a New York-specific item similar to type of reporting source), and the date of first contact. The algorithm was evaluated by comparing results to those obtained with manual review by experienced certified tumor registrars (CTRs).

Results: From a sample of 209,907 tumors with multiple diagnosis dates reported to the New York State Cancer Registry (NYSCR), the algorithm determined a single date for 94.7 percent of these, with the balance designated for manual review. Of a sample of 636 tumors that were manually reviewed to evaluate the algorithm, the algorithm obtained the same year as the CTRs for 621 tumors (97.6 percent), the same month and year for 572 tumors (89.9 percent) and the same month, year, and day for 518 tumors (81.4 percent). There was much lower agreement between the manually derived dates and the originally consolidated dates.

Conclusion: The algorithm presented here is accurate, efficient, and reliable, and hopefully will help the cancer registry community move toward standard practices for record consolidation.