Representativeness of an HIV cohort of the sites from which it is recruiting: results from the Ontario HIV Treatment Network (OHTN) cohort study

BACKGROUND: Participation bias is a well-known phenomenon in epidemiologic research, where individuals consenting to research studies differ from individuals who are not able or willing to participate. These dissimilarities may limit the generalizability of results of research studies. Quantification of the participation bias is essential for the interpretation of research findings. METHODS: The Ontario HIV Treatment Network Cohort Study (OCS) is an ongoing open cohort study of HIV positive individuals receiving care at one of 11 sites in Ontario. OCS participants from 4 sites were compared to non-participants (those who declined or were not approached) at those sites with regard to gender, age, HIV risk factor, CD4 count and viral load (VL). Generalized logit regression models were used to identify predictors of declining to participate or not being approached to participate. RESULTS: Compared to participants (P) in the OCS, individuals who declined to participate (D) and those who were not approached (NA) were slightly younger (D:45, NA:44 vs P:46), less likely to be male (D: 71%, NA:75% vs P:88%), less likely to be Caucasian (D:41%, NA:57% vs P:72%) and less likely to be Canadian-born (D: 39%, NA: 52% vs P: 69%). Patients who were not approached to participate were less likely to have VL < 50 copies/mL than other patients (D: 75%, NA: 62%, P: 74%) and had lower CD4 counts than OCS participants (D: 450 cells/mm(3), NA: 420 cells/mm(3), P: 480 cells/mm(3)). CONCLUSIONS: Significant demographic and clinical differences were found between OCS participants and non-participants. Extrapolation of research findings to other populations should be undertaken cautiously.