Death Clearance Process Improvement

Missouri Cancer Registry

To meet National Program of Cancer Registries (NPCR) requirements and North American Association of Central Cancer Registries (NAACCR) Gold standards, a central cancer registry (CCR) must follow death clearance guidelines and produce an annual incidence dataset that contains no more than 3% death certificate only (DCO) cases.

The Missouri Cancer Registry (MCR) initiated death clearance and follow-back in 1996, the registry’s NPCR reference year. For diagnosis years (DY) 1998 through 2007, MCR met the minimum requirement of less than or equal to 5% DCO cases. During DY 1999, MCR met the higher standard of less than or equal to 3% DCO cases. However, the December 2008 data submission for DY 2006 showed an increase from the two preceding submissions (2.49% DCO vs. 2.28% for DY 2005 and 2.14% for DY 2004). After receiving the 2007 death file from the Missouri Department of Health and Senior Services (DHSS) Bureau of Vital Statistics, MCR staff realized that the DCO percentage for DY 2007 might approach 3%, since the file included 165 cases marked for follow-back, but no follow-back source was available—there was no facility to be contacted, and the physician field contained the equivalent of "unknown."

To avoid an increase in the percentage of DCO cases in MCR's December 2009 data submission to NPCR, we decided to do a pilot linkage with DHSS's Patient Abstract System (PAS) database. MCR staff created a file containing the names of the 165 Missouri residents who died in 2007, but whose death certificate did not contain a follow-back source. DHSS staff linked that file to their PAS database. The result was a file that contained hospital admissions from 2002–2007, along with as many as 22 diagnosis codes for each of the 137 death-patient matches.

MCR staff reviewed the International Classification of Diseases, Ninth Revision (ICD-9) and Current Procedural Terminology (CPT) codes on all matches to identify the first time any type of cancer code was cited. If the PAS information verified the cancer, quality assurance staff used the first date the cancer was coded to change these records from DCO cases to hospital cases. Of the original 165 patients, the PAS linkage matched 137 (83%). Of the 137 deaths with one or more hospital admissions, MCR staff cleared the DCO status for 82 cases (60%)—about half of the total number of deaths sent for matching. The most common primary site for cleared cases was lung cancer (33%), followed by hematopoietic cancers (13%); prostate and pancreatic cancer accounted for 11% and 6%, respectively.

As a result of the PAS linkage and other follow-back strategies, MCR’s DY 2007 dataset was submitted to CDC by December 1, 2009, and contained only 2.16% DCO cases.

We learned—

• Cancer registries benefit from having an employee familiar with ICD-9 and CPT codes to review the initial death file-PAS matches.
• The majority of the 137 cases identified through the linkage were coded to a specific cancer diagnosis, rather than to "history of" a particular cancer (which accounted for only 5% of death certificate-PAS matches).
• Many cases were for cancers that may be diagnosed through radiology or cytology rather than by pathology.
• If a patient has many admissions for a variety of conditions over several years, it is often possible to pinpoint the admission with the cancer diagnosis.
• For patients with only one or two admissions, the cancer was generally diagnosed during the first admission.

The pilot project had several limitations. First, it had to be completed in a short time frame. Consequently, we limited linkage to only those deaths where no follow-back source was available and only tried to link with hospital admissions from 2002 through 2007. When the linked file was returned to MCR, it did not contain the requested hospital identifier; consequently, follow-back to admitting facilities, which might have cleared more cases and provided additional information on linked cases, could not be carried out.

For DY 2008 cases, we plan to start the process earlier, expand linkage to include cases where a follow-back source is available but no or insufficient information can be obtained, and obtain hospital identifiers. We will explore the feasibility of linking with admissions prior to 2002 and of using PAS linkage to obtain missing data such as Social Security number and race.