Validating a clinical prediction rule for chlamydia and gonorrhea infection among online testers in British Columbia, Canada
Research theme(s)
Internet Based Testing
Aidan Ablona, Titilola Falasinnu, Michael Irvine, Devon Haag, Hsiu-Ju Chang, Sharmistha Mishra, Ann Burchell, Christopher Fairley, Nathan Lachowsky, Paul Flowers, Claudia Estacourt, Michelle Murti, Oralia Gomez-Ramirez, Mel Krajden, Troy Grennan, Mark Gilbert
STI & HIV World Congress, July 14-17, 2019
Background
Clinical prediction rules (CPRs) estimate the probability of a health outcome to support decision-making in intervention and service delivery. Previously, a CPR was derived to maximize detection of chlamydia and/or gonorrhea (CT/GC) infections and minimize the number of screening tests offered among asymptomatic women and heterosexual men attending sexually-transmitted infection (STI) clinics in Vancouver, British Columbia, Canada. We assessed the external validity of using this clinic-derived CPR within GetCheckedOnline (GCO), a provincial online STI testing program in British Columbia.
Methods
Data used for calculating CPR scores, including age, race/ethnicity, number of sexual partners and previous CT/GC diagnoses, were collected prospectively on GCO from October 2015 to June 2018. Model calibration and discrimination were evaluated using the Hosmer-Lemeshow (H-L) statistic and the area under the receiver operating characteristic curve (AUC), respectively. Sensitivity and proportion of GCO clients screened were calculated at different CPR cut-off scores. In the original derivation population, the CPR had an AUC=0.74, with a cut-off risk score ≥6 identifying 91% of infections and screening 68% of testers.
Results
Among 2703 GCO CT/GC test episodes, the prevalence of CT/GC infection was 2.1%. The clinic CPR showed reasonable calibration (H-L p=0.952) and discrimination (AUC=0.64, 95%CI: 0.57-0.71). Using a CPR cut-off risk score of ≥6, we would have detected 79% of infections and screened 64% of testers. Lowering the cut-off risk score to ≥4 would have increased sensitivity to 95% while screening 85% of testers.
Conclusion
This is the first study validating the use of a clinic-derived CPR within an online setting. Our CPR showed reasonable accuracy and performance when applied to GCO data. Differences in model performance online compared with clinic-based settings highlight important differences in the populations who use online testing. Use of CPRs in online contexts offers unique and novel opportunities for public health and STI testing.