Sample collection points Totally, 169 fluid milk samples were collected from different entry points in Chittagong city in Bangladesh including, Shikalbaha, Sholoshahar Railway Station, Jalalabad Market, Chittagong city Gate, Halishahar, Chittagong Port, and Chittagong Batali Road. Also, 17 samples were directly collected from a dairy farm located in Shikalbaha. All the samples were collected from September 2013 to March 2014.
Data collection and survey method A questionnaire designed for this study was pretested according to Yien (2014) with some modifications. The questionnaire was then administered by two trained veterinarians who interviewed milk vendors at time of milk collecting. The main assessed variables during interviews were geographic location, amount of milk collected, stock information, flock health history, and overall farm management.
Identification of E. coli For initial screening of E. coli, the collected samples were prepared by serial dilution. At first, 100 µl from each milk sample was transferred to 900 µl sterile peptone water (0.1%) and thoroughly mixed to give 1:10 dilution and then the serial dilutions were prepared. Each 100 µl diluted milk sample was inoculated onto MacConkey agar medium (Oxoid, Basingstoke, Hampshire, UK), incubated for 24 h at 37 °C. The colony counts of large pink color colonies on medium were presented as Colony Forming Unit per ml (CFU/ml). At the same time, diluted milk samples were inoculated onto blood agar medium (Oxoid, Basingstoke, Hampshire, UK) enriched with 5% defibrinated bovine blood for total plate count by colony counter. Five large pink color colonies from MacConkey agar medium were homogenized and inoculated onto an eosin methylene blue (Oxoid, Basingstoke, Hampshire, UK) agar plate, incubated at 37 °C for 24 h. Convectional biochemical tests were performed for identification of E. coli isolates using triple sugar iron agar, urea agar, methyl red broth, and simmon citrate agar.
Statistical analysis The data were analyzed by Stata 11 (Stata Corp, College Station, Texas, USA). P value <0.05 in two-sided Fisher’s exact test was considered as significant. Any variable with p<0.20 was considered for multivariable analysis. Logistic regression was applied for multivariable analysis. A backward stepwise variable selection strategy was used to construct a final model with a significance level of p<0.05.