Discussion

Our findings suggest that the risk factors traditionally used for colorectal surgery SSI surveillance in the European setting account for ∼60% of its incidence. Ergo, a substantial proportion of SSI incidence is explained by variables that are not present in the current methodology. Modifiable risk factors may be responsible for at least 31% of SSI incidence. Of the 3 strongest risk factors with highest impact on SSI following colorectal surgery, 2 factors were potentially modifiable. Interventions focusing on adopting techniques that are less invasive and that provide shorter surgeries may be optimally suited for prevention. The use of laparoscopy has been steadily increasing in recent years, both nationally (Table 2) and internationally, ^{Reference Ghadban, Reeh and Bockhorn25,Reference Stormark, Søreide and Søreide26} and its progressive introduction has been shown to work as a protective factor not only for overall and incisional SSI but also for organ-space SSI. ^{Reference Arroyo-Garcia, Badia and Vázquez27} These results suggest that although these risk factors cannot be entirely eliminated, there is still a considerable margin to decrease their prevalence and, thus, their impact on SSI incidence. Nonmodifiable variables estimates are mainly dependent on the wound classification, in which the estimate has low precision. Urgent operations and lack of antibiotic prophylaxis, although relevant risk factors for SSI, had the lowest PAFs due to their low prevalence.

Our results suggest that unmeasured risk factors may be responsible for 40% of incidence, a high enough value to warrant close attention. The HAI-Net protocol determines the collection of the same variables for all surgical procedures to guarantee comparability, yet risk factors may be surgery specific. The closing technique of ileocolic anastomosis have been shown to be associated with SSI and could be relevant to monitor, even if it applies solely to colorectal surgery. In contrast to most other surgery types, SSI after colorectal procedures is mainly attributed to endogenous gut bacteria. ³ Therefore, all operative factors associated with more invasive procedures or delayed wound healing are expected to have a considerable impact on SSI incidence and could be considered. Comorbidities are also not directly included in the HAI-Net protocol. Even if comorbidities are unlikely to be acutely modifiable prior to surgery, knowing their impact would help to ascertain the acceptable baseline level for each setting, considering the patient case mix. Nonetheless, most comorbidities (ie, obesity, smoking habits, alcohol use, diabetes) are indirectly expressed in the ASA classification. ^{Reference Doyle, Goyal and Bansal12} Comorbidities such as diabetes, which are usually studied in a dichotomous manner, ^{Reference Xu, Qu and Gong13} are evaluated as well-controlled factors or are not in the preoperative assessment and, thus, better represent the overall status of the patient. Because many comorbidities tend to be present in the same patients, an adjusted PAF would most likely yield low estimates for each individual factor. A weighted PAF of 6% for the ASA score may suggest that regardless of the how comprehensive a surveillance system may be on the patient’s medical history, its impact on SSI incidence may be low. Even so, patient and procedural variables may not tell the full story. Some part of the missing picture may be attributable to hospital characteristics, such as availability of surveillance, or surgeon caseload volume. However, robust studies providing accurate estimates are still lacking. ^{Reference Malheiro, Peleteiro and Correia28} Nevertheless, compliance with prevention care bundles may be the most relevant missing variable. Care bundles consist of a set of simple, strong evidence-based practices that, when implemented in a combined and consistent manner, improve patient outcomes, with a higher impact than the addition of each intervention effect. ²⁹ Although they have been shown to decrease SSI after colorectal surgery up to 45%, their success is deeply dependent on compliance with the entire bundle. ^{Reference Zywot, Lau, Fletcher and Paul30,Reference Pop-Vicas, Abad, Baubie, Osman, Heise and Safdar31} Suboptimal compliance rates may justify Portugal’s place in SSI incidence at the European level and are strong candidates for the unexplained remaining incidence we found. ³

Our secondary analysis of hospitals participating in Stop Infeção Hospitalar! showed that, for engaged hospitals, measured variables explain a lesser proportion of infection. Surveillance may decrease SSI rates through feedback and surveillance effect, which may justify why participating hospitals, with more robust surveillance teams, have a higher proportion of infection attributed to unmeasured risk factors. ^{Reference Gastmeier, Schwab, Sohr, Behnke and Geffers32} Nonetheless, the difference was not significant. Although we acknowledge that no surveillance network can include every single risk factor for infection, it is vital to understand which variables are relevant enough to warrant inclusion in future updates of the protocol.

To the best of our knowledge, this is the first study to address the impact of risk factors in the context of SSI in colorectal surgery and the first in the infection prevention and control field to adjust estimates for nonindependence of risk factors. The weighted-sum approach also allows the estimation of a combined PAF that is interpretable as a measure of the combined impact of the risk factors considered, which is a known limitation of unadjusted impact measures. Although it has been described previously as a method of adjusting PAF for confounding, the ideal weight remains a matter of discussion. ^{Reference Benichou20,Reference Darrow and Steenland33} The use of communalities is a novel solution that has only recently been used in the field of Alzheimer’s disease and dementia. ^{Reference Mukadam23,Reference Ma’u, Cullum, Cheung, Livingston and Mukadam24} It allows researchers to account for interaction between risk factors and provides a more robust, even if conservative, estimate than simply applying adjusted relative risk using the Levin or Miettinen formula, both of which have been shown to yield tremendous bias. ^{Reference Benichou20,Reference Darrow and Steenland33} In our analysis, there was no obvious heterogeneity of variables, with all variables sharing some variance with the others. No variable was redundant, and all were usually defined by a communality >90%. ^{Reference Tabachnick and Fidell34}

Our sample size, the use of national data following the European methodology of surveillance, and the use of measures of association published in peer-reviewed meta-analysis are study characteristics that strengthen the external validity of our findings. Nonetheless, variations are expected in different settings. Prevalence estimates are setting dependent and may justify a differential impact elsewhere. Open surgery is still more prevalent in our setting than in other European countries, ^{Reference Ghadban, Reeh and Bockhorn25,Reference Stormark, Søreide and Søreide26} and the same is true for the ASA score. ^{Reference Hennessey, Burke, Ni-Dhonochu, Shields, Winter and Mealy35}

Notably, hospitals are not required to provide data for all performed procedures. Hence, potential selection bias may occur for procedures registered in national databases. Likewise, the variance overlap was calculated for 83% of our sample, which could have affected its representativeness. Most of the excluded procedures were missing data on antibiotic prophylaxis due to the changes in the data set outlined earlier. Although missing procedures were demographically similar to included procedures, differences in proxies of infection, such as the ASA score and the operation time, meant that included procedures had a higher risk of infection. We were unable to determine whether this had any influence in the values estimated. Nonetheless, our sensitivity analysis provided communalities in approximately the same relative position for each risk factor, with only minor changes in weighted PAF values. Other missing data, such as the proportion of laparoscopic surgeries that ended up as open surgeries, were unavailable.

In conclusion, variables routinely collected under the ECDC HAI-Net protocol explain 60% of SSI incidence after colorectal surgery, meaning that a relevant proportion of SSIs remains unaccounted for by current surveillance indicators. Interventions focusing on shorter, less invasive procedures may be optimally effective in reducing the burden of infection.