Table 2.
Optimal feature subset in final multivariable mixed-effects random forest model obtained through the Recursive Feature Elimination algorithm with antimicrobial usage in TIDDDvet as outcome
| Features | Count | Gini importance index | Effect size (95% CI) | Sub-block (Biological expectation) |
|---|---|---|---|---|
| Average number of chicks per round (null model; ln transformed) | 93.0 | −0.046 (−0.053 to −0.038) | Null model (+) | |
| Average number of rounds per year (null model) | 77.1 | −0.126 (−0.132 to −0.12) | Null model (+) | |
| Number of people working at the poultry farm in total (null model; ln transformed) | 72.4 | −0.118 (−0.128 to −0.108) | Null model (−) | |
| Are visitors obliged to check in before having entrance to the stables? 0 – no 1 – yes |
9 | 37.2 | 0.488 (0.458–0.518) | External (−) |
| In how many steps is the poultry house depopulated? 0 – In one step 1 – In two or more |
7 | 28.9 | 0.307 (0.294–0.32) | External (+) |
| Is there a farm specific protocol of non-officially controlled disease that is complied with? 0 – no 1 – yes |
6 | 28.7 | 0.51 (0.469–0.55) | Internal (−) |
| Is drinking water system fully disconnected and cleaned/disinfected? 0 – not after every round 1 – after every round |
9 | 22.2 | −0.519 (−0.547 to −0.49) | Internal (−) |
| Are there disinfection baths for vehicles present at the entry of the company? 0 – No 1 – Yes |
7 | 20.3 | 0.225 (0.21–0.24) | Internal (−) |
| Length ventilation is used in the sampled barn 0 – No 1 – Yes |
7 | 19.8 | 0.318 (0.304–0.332) | External (NA) |
| What is the poultry density in the sampled house? 0 – ≤ 37 kg/m2 1 – ≥ 38 kg/m2 |
7 | 19.5 | −0.048 (−0.058 to −0.038) | Internal (+) |
| Is the farm fenced-off? 0 – not completely 1 – completely |
7 | 18.6 | 0.157 (0.149–0.166) | External (−) |
| Are specific preventative measures taken for material supply? 0 – no 1 – yes |
8 | 15.9 | −0.205 (−0.216 to −0.195) | External (−) |
| Is stable-specific clothing available? 0 – no 1 – yes |
7 | 14.1 | −0.109 (−0.115 to −0.103) | Internal (−) |
| Is efficacy of cleaning/disinfection checked by taking a hygienogram? 0 – never 1 – not never |
6 | 13.4 | 0.068 (0.062–0.075) | Internal (−) |
| Is material stored per stable and recognisable? 0 – no 1 – yes |
7 | 12.7 | −0.058 (−0.065 to −0.051) | Internal (−) |
| Are visitors/personnel obliged to wash/disinfect hands before entering? 0 – no 1 – yes |
6 | 12.5 | 0.141 (0.134–0.148) | External (−) |
| Is there a FARM-hygiene lock present? 0 – No 1 – Yes |
6 | 12.1 | −0.071 (−0.078 to −0.065) | Internal (−) |
| Allowed for visitors/traders to enter stables/have direct contact? 0 – never 1 – not never |
6 | 11.8 | −0.091 (−0.1 to −0.082) | External (+) |
| When checking water quality, where is the water sample taken? 0 – not at nipple 1 – at nipple |
7 | 11.5 | −0.095 (−0.101 to −0.088) | External (−) |
| Roof ventilation is used in the sampled barn 0 – No 1 – Yes |
7 | 11.0 | −0.101 (−0.107 to −0.096) | External (NA) |
| Is there natural stagnant or running water within less than 1 km of farm? 0 – no 1 – yes |
6 | 10.6 | −0.006 (−0.011 to −0.002) | External (+) |
The features are ordered by Gini Importance index. In the fourth and fifth columns in bold are the variables for which their effect size sign and biological expectation of the measure match and thus they satisfy one of the criteria for being selected as candidate intervention on the farm.