A FUZZY PATIENT RECORD MATCHING ALGORITHM BASED ON HIV PATIENT DATA
MetadataShow full item record
Background: Accurate patient information including transfer-in and outs at every hospital visit is crucial for the successful HIV care and treatment. Patient matching across different health facilities is still a challenge due to the lack of universal unique identifiers. Although attempts to use identifier techniques like fingerprint identification and National ID have been implemented, there still remains several challenges. These include the fact that not all people have National IDs and the use of finger print technologies is prone to many technical failures which affect record matching. Fuzzy matching involves using patient data to determine relations in two or more datasets. This approach provides a possible solution in absence of universal unique identifiers, enabling matching of data from different HIV clinic databases. We aimed to develop and test a fuzzy matching algorithm, based on a set of parameters from information commonly collected by various HIV health facilities. Methods: This was an experimental study at HIV/AIDS clinics in five KCCA health facilities. Patient records were obtained and we designed two data cleaning algorithms and a Fuzzy Patient Matching algorithm which we also implemented in OpenMRS. We then evaluated the effectiveness of the three algorithms using precision, recall and F-score. For the FPM algorithm, effectiveness was tested using experiment 1 and experiment 2 for both set 1 and set 2 of the variable weights of 14 combinations of patient variables. In experiment 3, we tested for a situation when the patient variables have been misspelled or may be missing. User satisfaction evaluation was conducted among three categories of OpenMRS cadres/users at the health facilities and the results analyzed in Stata 14. Results: The fuzzy record matching variables were found to be name, address, gender, birthdate, telephone number, treatment supporter telephone number, ART start date and first encounter date but a minimum of 10 out of 14 variables was recommended to ensure a higher accuracy in matching. The Data cleaning algorithms achieved precision and recall of 100%, and an accuracy above 90%. In experiment 1, 2, and 3 the FPM algorithm achieved precision and recall above 100%. Furthermore, 81.9% found the tool useful, 86.4% found it easy to use,90.9 found it easy to learn, 85.5% were satisfied with the tool and 95% found the tool usable. Conclusion: The FPM algorithm offers technique to match patient records across different facilities in the absence of a universal unique identifiers. Therefore, as the Ministry of Health rolls out fingerprint identification and use of national identification numbers, it is important to note that in circumstances where fingerprint technology fails or where it cannot be implemented, fuzzy patient matching can be used.