The Unreliable Identifying Key Problem

Hidden keys allow the central registry to detect when a hospital registry has updated other patient identifiers. For example when the social security number (SSN) and last name of a previously submitted patient are corrected at the hospital and resubmitted to the central registry, how does the central registry detect that this is an update and not a new patient?

123-45-6789 John Smith = 223-45-6789 John Smithe

With hidden keys in place, the software identifies the previously submitted recrods using the hidden keys and thus detects that an update was performed. i.e. the hidden keys will match, even though the other identifying keys do not.

Regrettably, NAACCR records do not contain reliable hidden keys. While space is allocated in the NAACCR record for such keys, experience has shown that hospital registry software packages besides CPDMS do not output them. In order to merge NAACCR records into a CPDMS database, it is necessary to generate hidden keys for the NAACCR records. A set of algorithms have been developed to perform this task. This is a particularly difficult objective because we must rely on traditional patient identifiers to retain patient identity even while the patient identifiers are potentially changing. This process is by definitition imperfect and the possibility exists that when multiple identifiers for a single patient are modified simultaneously, then the software may conclude, incorrectly, that the incoming record is a new patient. However, great effort has been made to to minimize this possibility. On the positive side, our task is limited to maintaining hidden keys for individual hospital registries. In other words, we need only compare identifiers of patients against records submitted by the same hospital.

Much research has been devoted to the process of Probabilistic Record Linkage. This is a process of utilizing probability theory to decide whether or not a set of identifiers uniquely identify each individual. Some probablistic methods have been incorporated into this process for CPDMS but it is not yet truly probabilistic. Probabalistic linkage would involve performing frequency analysis on individual identifiers and assigning weights to matching values. It is noted that commerical software such as MatchWare could be used to search for duplicates in the CPDMS database.

Ulitimately this problem would be resolved if hospital based registry software could generate hidden keys and output them in the NAACCR record. Alternatively, it would be resolved if they could incorporate hidden keys generated by the central registry into the hospital record and output them with NAACCR records. Until this occurs, we will rely on the following algorithms.

Patient Identifying Keys Utilized by CPDMS NAACCR Merge

If Hospital Accession Year and Number Match {
	If SSN Matches, Accept/Overwrite Patient Keys.
	If Last Name and First Name Match, Accept/Overwrite Patient Keys.
	Otherwise Reject for Review.
}

If SSN Matches {
	If Last Name, First Name Match, Accept/Overwrite Patient Keys.
	If NYSIIS Last, NYSIIS First, Sex & BirthDate, Match, Accept/Overwrite.
	Otherwise Reject for Review.
}

If (Name and/or NYSIIS), Sex/BDate, Address, & Phone Match, Accept/Overwrite Record.
Otherwise, If Any 2 Match, Reject and Report for Review;
Otherwise, If Fewer Than 2 Match, Accept New Record.