75716

COMPARING THE ACCURACY OF LINKAGE TOOLS TO RETRIEVE VITAL STATUS DATA FROM PLWHA DATABASES

Favoritar este trabalho

Objective: To compare the accuracy for two linkage routines utilized to retrieve vital status information from people living with HIV/AIDS (PLWHA) databases. Methods: Nominally-identified records from PLWHA aged 13 were obtained from SIM, SICLOM and SISCEL. Linkage routines included a previously validated fully automated algorithm in Python language (Py) and Reclink (Rl), a probabilistic software largely utilized. Records from PLWHA known to be alive (in AIDS care by the end of 2012) were added to records from patients reported as deceased in SICLOM between January 2008/December 2009. Data were then searched in the SIM database over January 2008 and December 2010. Scenarios where 5% and 50% of patients were actually dead (PAD) were simulated, considering both complete cases (CC) and 20% missing maternal names (MMN). Rl parameters followed Health Ministry recommendations. Results: Sensitivities for Py/Rl were 96.8% (94.2 99.3)/98.4% (96.6/100.0) respectively, when CC were available and 94.6% (91.3-97.9)/94.6% (91.3-97.9) in MNM, whereas specificities were respectively, 99.0% (97.6-100.0)/99.5% (98.5-100.0) and 99.9% (99.8-100.0)/99.6% (99.4-99.8), p <0.01 for the latter. Positive predictive values/negative predict values (PPV/NPV) for both routines had similar behavior, ranging from 96.8% (94.3 99.3) to 99.9% (99.8 100.0), except for the PPV in a 5%PAD MNM scenario, in which the values were 97.8% (95.6-99.9) and 91.6% (87.7-95.6, p<0.01) for Py and Rl, respectively. Manual reviewing in Rl ranged from 6% to 7% of the records. Conclusions: While both linkage routines seem to be interchangeable in retrieving vital status data from PLWHA, fully automated Py draws particular attention as it spares manual subjective analyses.