Design and participants
The SIF in Vancouver is being evaluated through the Scientific Evaluation of Supervised Injection (SEOSI) cohort, which has been described in detail [21]. Briefly, the cohort was assembled through random recruitment of IDU from within the SIF. Random recruitment is based on inviting users of the SIF to be referred to the research study during random blocks of time. Among individuals who were recruited, a venous blood sample was drawn and an interviewer-administered questionnaire was conducted at baseline and at semi-annual follow-up visits. The informed consent agreement, obtained for all participants, included a request to link the SIF evaluation with administrative health databases. In Vancouver, hospitals are equipped with a database that tracks patient admission. The SIF is also equipped with a similar database. At the SIF, nursing care includes wound care, dressing changes, and measuring temperature. However, there are no physicians at the SIF. If microbiological investigation or medical treatment, including antibiotic prescription or intravenous antibiotic therapy is required, individuals at the SIF must be referred to the hospital or a medical clinic. In this study, a linkage of SEOSI participant data, SIF data and St. Paul's Hospital inpatient data was performed. St. Paul's Hospital is the major urban hospital serving the DTES community, one of Canada's poorest postal codes. The University of British Columbia-Providence Health Care Research Ethics Board approved the present study.
Measurements
The start point for these analyses was enrollment into the SEOSI cohort and the endpoint was hospitalization for a CIRI or related infectious complication. The infectious complications included were based on previous literature [8, 9]. The definition of the reason for hospitalization was based on International Classification of Diseases (ICD) 10 codes on patients' hospital records and included: abscess (G061, G062, L020, L021, L022, L024, J851), cellulitis (L0300, L0310, L0311, L032, L0335, L038), osteomyelitis (M4620, M4625, M4629, M8617, M8618, M8661, M8663, M8666, M8681, M8691, M8695), staphylococcal infection 161 {(A490, A499, B956) including, septicaemia (A410, A412, A419) and Methicillin-Resistant Staphylococcus aureus (MRSA), (U000)}, endocarditis (I330), septic arthritis (M0000, M0002, M0004, M0005, M0006, M0008, M0009), ulcer (L089, L979), thrombophlebitis (I802, I808) and myositis (M6005, M6008). A few ICD-10 codes chosen suggest but do not require infectious etiologies (e.g., thrombophlebitis, myositis, and ulcer). The number of events refers to a CIRI or related infectious complication as a primary, secondary, tertiary, quaternary, or quinary diagnoses according to relevant ICD 10 codes unless otherwise specified.
We first examined the distribution and frequency of a CIRI or related infectious complication. Then, we evaluated length of stay in hospital among study participants and examined this outcome as a continuous variable in a linear model that adjusted for the following confounding variables: age, sex, HIV serostatus and SIF nurse referral. We then considered the cost of hospitalization, associated with CIRI, which was estimated at C$712 per hospital day, based on a fully-allocated costing model for the province of British Columbia from 2001 [1]. This estimate was updated to Canadian dollars in 2005 [22]. Fully-allocated costing includes costs associated with nursing care, medications, investigations, physician visits and length of stay as well as overhead, opportunity cost of hospital resources and a 5% depreciation of capital equipment [1]. Potential healthcare savings were estimated by multiplying the cost per day value (C$712) by the difference in number of days hospitalized among individuals with CIRI who were referred by a nurse within the SIF and those who were self referred to the hospital.
We investigated baseline characteristics stratified by hospitalization or not bivariately. Using Cox proportional hazard regression, we examined factors potentially associated with hospitalization. Variables considered for our analyses included: age; sex at birth (female vs. male); current residence in DTES (yes vs. no); living in unstable housing (yes vs. no); daily cocaine injection (yes vs. no); daily heroin injection (yes vs. no); daily speedball injection (yes vs. no); and HIV serostatus (positive vs. negative). As used previously, unstable housing was defined as living in a single room occupancy (SRO) hotel, shelter, recovery or transition house, jail, on the street, or having no fixed address [23]. Variables from the semi-annual questionnaire referred to behaviour that occurred in the last six months unless otherwise specified. We also examined whether a SIF nurse referral to the hospital was associated with hospitalization and, if so, whether length of stay was different given referral versus self referral. For this task, we conducted a record linkage matching their SEOSI identifying code to each participant's record in the SIF database to determine nurse referral. Then, we linked his or her SEOSI identifying code with his or her unique personal health number to examine hospital records prior to the censor or event date.
Variable selection was based on previously published literature on a CIRI or related infectious complication and hospitalization among IDU [1, 5, 8, 9, 20]. Variables considered associated with hospitalization were analyzed in unadjusted analyses and adjusted Cox proportional hazard regression model. Time zero was defined as the date of recruitment into the SEOSI study for all participants and participants not hospitalized at St. Paul's Hospital were censored as of 31 January 2008. All behavioral variables were treated as time-updated covariates based on semi-annual follow-up data. The multivariate model was fit using a fixed model whereby we included all variables that were statistically significant at the p < 0.05 level in univariate analyses. All statistical analyses were performed using SAS 8.0 (Cary, NC) and all p-values were two-sided.