A Keyword Search Strategy to Identify Missed Cases of Myelodysplastic Syndromes in Population-Based Cancer Registries.

Myelodysplastic syndromes (MDS) were reclassified from blood disorders to neoplasms in the tenth edition of the International Classification of Diseases and, as a result, became reportable malignancies to population-based cancer registries in 2001. Recent analyses of data from the North American Association of Central Cancer Registries (NAACCR), which includes registries reporting to the National Cancer Institute's Surveillance, Epidemiology, and End

(SEER) Program, provided the first opportunity to investigate the incidence and survival of patients with myelodysplastic syndromes (MDS) in the U.S. However, several lines of evidence suggest that reported MDS incidence rates are considerably underestimated (Rollison, Blood 2008). Due to the unique patterns in diagnosis and treatment of MDS, many MDS patients may not access hospital-based care, particularly during the early stages of their disease. These cases are potentially missed by population-based cancer registries if they are not routinely reported to such registries by their private physicians. Given the potential for under-reporting, it was hypothesized that the true incidence of MDS is higher than currently estimated by population-based cancer registries and that previously missed MDS cases could be identified through careful systematic review of electronic pathology reports obtained from private laboratories. To test this hypothesis, a feasibility pilot study was initiated in collaboration with the Florida Cancer Data System (FCDS), the statewide cancer registry, which uses electronic pathology (E-Path) reporting.

All E-Path reports sent by private pathology laboratories to FCDS in 2006 were queried using MDS keyword terms, including words and phrases potentially representative of MDS (e.g., myelodysplastic, ringed sideroblast, Pelger-Huet, etc.). E-path reports that matched one or more of the search terms were compared to the FCDS database to distinguish E-path reports that corresponded to individuals already in the FCDS database from those that corresponded to individuals who were not in the FCDS database. For those individuals within the FCDS database that linked to one or more E-path reports, demographic characteristics were compared between those with a previous MDS diagnosis recorded in MDS and those with one or more diagnoses of cancers other than MDS. Within the latter group, E-path reports were categorized by number of keyword hits, and a random sample of 50 E-path reports from each category were reviewed by a single hematologist/oncologist (CRC) to confirm the diagnosis of MDS. The percentage of missed cases was calculated as the number of E-path reports that were determined to be MDS divided by the number of E-path reports reviewed.

The initial query captured 121,279 E-path reports. After excluding 40,894 duplicate records, 80,385 unique E-path reports were identified, of which 19,812 linked to a cancer patient registered in FCDS. Of those 19,812 E-path reports, 1,452 (7%) linked to patients for whom a diagnosis of MDS was recorded in FCDS, and 18,357 linked to patients with cancer diagnoses other than MDS. The probability of an E-path report linking to an MDS case increased with the number of keyword hits in the E-path report (p <0.0001). As compared to FCDS-registered patients with cancers other than MDS who linked to an E-path report matching at least one MDS keyword, those registered with MDS were older (p<0.0001) and more likely to be male (p = 0.0002). Based on the review of 200 randomly selected cases, the overall percentage of missed MDS cases was 3.5%, with the percentage increasing with number of keyword hits. For reports deemed non-MDS by the cancer registry yet matching 6+ MDS keywords, at least 14% were missed cases of MDS.

This pilot study demonstrated the potential for MDS cases to be missed, even when the patients are already registered as having another type of cancer in population-based cancer registries. Application of a keyword search strategy to identify missed cases of MDS among electronic pathology reports is a feasible technique for improving case ascertainment of MDS in population-based cancer registries. Given the existence of missed MDS cases, it is likely that MDS incidence rates are underestimated at the population level.

No relevant conflicts of interest to declare.