308
nosocomiales (SPIN; the provincial nosocomial infection surveil- lance program) required participating hospitals to perform active HABSI surveillance in their facility, excluding psychiatric wards, long-term care, and nurseries. On April 1, 2013, participation in BACTOT became mandatory for all hospitals with >1,000 admis- sions per surveillance year. A surveillance year begins on April 1 of a calendar year, ends on March 31 of the following calendar year, and is composed of thirteen 4-week administrative periods. For every period, the following data are collected for each facility: total patient days, patient days in the ICU, and all relevant information on identified HABSI cases.
Case definitions
The BSI case definition was described elsewhere24 and was based on the National Healthcare Safety Network criteria.25 To be con- sidered healthcare-associated, a BSI could not be present or incu- bating within 48 hours of admission, except if it resulted from a preceding admission or procedure. Primary BSIs constitute BSIs associated with a venous catheter (CA-BSI), both central or periph- eral, and non–catheter-associated primary BSIs (NCA-BSI). Secondary BSIs followed by BACTOT are those arising from pri- mary surgical site infections (BSI-SSIs), urinary tract infections (BSI-UTIs), pulmonary infections (BSI-PULMs), intra-abdominal infections (BSI-ABDOs), skin and soft-tissue infections (BSI- SSTs), bone and joint infections (BSI-BONEs), or any other infec- tion of primary focus (BSIOther).
Study design and analysis
We conducted a secondary analysis of BACTOT, a retrospective cohort study using HABSI data collected for BACTOT and obtained directly from SPIN. The cohort was open and included hospitals that had participated for at least 3 consecutive surveil- lance years in BACTOT by the end of 2016–2017. This restriction allowed us to compare the year of entry rates with those of at least 2 subsequent surveillance years. Yearly participation was defined as contributing to at least 11 of the 13 periods within the surveillance year. Hospitals with no cases (n=2) were excluded because they would not contribute any information to the fitted models. Access to data was granted by Institut national de santé publique du Québec (National Institute of Public Health) to allow the authors to carry out its monitoring mandate entrusted by the Ministry of Health and Social Services. The McGill University Institutional Review Board approved this study.
Numerators All HABSIs among admitted patients were considered cases. Cases were pooled by hospital, administrative period, and surveillance year and were stratified by type of infection.
Denominators Patient days were pooled by hospital, administrative period, and surveillance year. Every day spent at a participating hospital by a patient was counted as 1 patient day. Days of admission and discharge were each counted as half
a day.
Descriptive analyses Hospitals thatmet the inclusion criteria were described by the num- ber of years they contributed, their teaching status, whether they had an ICUor not, and number of beds. The frequency distribution of HABSI cases by infection source over the 10-year period was
Iman Fakih et al
computed. Raw pooled HABSI rates per 10,000 patient days were calculated for each period by dividing the number of HABSI cases by the total number of patient days. The 95% confidence intervals (CIs) for these rateswere calculated using the normal approximation method. Percentages may not equal 100 because of rounding.
Statistical analyses Multilevel Poisson-lognormal mixture models were fitted to the data aggregated by hospital, period, and surveillance year with HABSI, CA-BSI, NCA-BSI, or BSI-UTI cases as the outcome and the natural logarithm of patient days as the offset. The remain- ing HABSI subtypes were too rare to achieve models with satisfac- tory fit, as evaluated by fitted values. The log-mean Poisson rate for each observation was decomposed into a surveillance (year) ran- dom component, a period random component, and a hospital component following a normal distribution with unknown vari- ance. The prior mean of the hospital component was modeled as a linear function of the number of beds in the facility and hos- pital type (nonteaching without ICU as the reference, nonteaching with ICU, and teaching), and a random effect relating to the year the hospital entered BACTOT. Independent zero mean normal prior distributions, with relatively large variance (=10) were assigned to the coefficients of the hospital-level variables. Some hospitals participated for <13 periods annually, and 10 observa- tions were missing. Patient days for those observations were imputed using multiple regression, with calendar time and number of admissions as covariates, while the outcome was considered to be missing completely at random. These missing observations were considered parameters of the model and were estimated imputed within the using Markov Chain Monte Carlo (MCMC) procedure. The surveillance effect was calculated for years 2–10 by expo-
nentiating their surveillance component and dividing it by that of the first year to get incidence rate ratios. Similarly, the period effects were calculated for periods 2 to 13 by exponentiating the period component and dividing it by that of the first period. The incidence rate ratio of different entry years was estimated by dividing the rate independently associated with the year of inter- est by that associated with the latest year of entry included in the study, 2014–2015. The sample from the posterior distributions of the coefficients of number of beds and hospital type were exponen- tiated to obtain incidence rate ratios. The same model was fitted using data from all hospitals and
then separately for 37 hospitals with <10 years of participation, for a total of 8 models. The subgroup analysis was performed to exclude hospitals that may have been conducting facility-wide HABSI surveillance prior to BACTOT entry, for those that started in 2007–2008. Because the analytical form of the posterior distribution of the
parameter vector is unknown, we used MCMC methods to obtain samples from the resultant posterior distribution.26 In particular, the models were fitted using JAGS software within the R package, rjags version 4–6.27,28 The model burn-ins were 60,000 followed by 150,000 sampling iterations. Convergence of the chains was checked using the Gelman-Rubin diagnostic.29 All analyses were conducted using R version 3.4.1 software with RStudio version 1.0.143 (RStudio Team, Boston, MA).
Results Cohort description
In total, 77 hospitals were included in the study, representing 87% of the hospitals eligible to participate in BACTOT. Among them,
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84 |
Page 85 |
Page 86 |
Page 87 |
Page 88 |
Page 89 |
Page 90 |
Page 91 |
Page 92 |
Page 93 |
Page 94 |
Page 95 |
Page 96 |
Page 97 |
Page 98 |
Page 99 |
Page 100 |
Page 101 |
Page 102 |
Page 103 |
Page 104 |
Page 105 |
Page 106 |
Page 107 |
Page 108 |
Page 109 |
Page 110 |
Page 111 |
Page 112 |
Page 113 |
Page 114 |
Page 115 |
Page 116 |
Page 117 |
Page 118 |
Page 119 |
Page 120 |
Page 121 |
Page 122 |
Page 123 |
Page 124 |
Page 125 |
Page 126 |
Page 127 |
Page 128 |
Page 129 |
Page 130 |
Page 131 |
Page 132
