Introduction

Several guidelines for the prevention of surgical-site infection (SSI) have been developed by the Centers for Disease Control and Prevention (CDC) [1, 2] and other agencies [3,4,5] in the United States. In the field of pediatric spinal deformity care, best practice guidelines for the prevention of SSI were developed based on expert consensus [6, 7]. There are ongoing efforts to build evidence in the field, and a review of recent literature reported that the use of antibiotic prophylaxis regimens, such as gentamicin-impregnated allograft bone and intra-wound vancomycin powder, could be effective in decreasing risk of SSI [8, 9]. However, the complex multi-factorial nature of SSI imposes challenges to identifying which individual patients will best benefit from implementation of specific preventive care factors. Precision prevention [10,11,12] ensuring that patients, especially those at highest risk, receive appropriate and targeted preventive care is important, particularly in the value-based model of healthcare which balances increasing quality of care and improving population health with decreasing per capita costs [13].

An SSI risk calculator was developed and tested to compute the probability of SSI in individual pediatric patients undergoing spine surgery [14]. The model used patient, surgical and hospital factors which are determined preoperatively are unlikely to be modified [14]. This calculator makes it possible to identify high-risk patients, enhances patient education and shared decision-making, target resources for preoperative optimization, and alerts the perioperative care team in advance. An essential next step is to identify modifiable preventive care measures to reduce the individual patient’s probability of SSI. Purposes of this study were to investigate the association between preventive care measures and patient risk of SSI, and to estimate the reduction of SSI probability by these preventive care measures from the baseline probability in individual patients, as calculated by previously identified risk factors [14]. We hypothesized that some preventive care factors would be associated with reduced incidences of SSI within 90 days after spinal surgery in individual patients.

Materials and methods

A multi-center retrospective study was conducted using data from seven institutions in urban hospitals with a broad range in number of surgical procedures (mean = 515, range = 8–1490). Upon IRB approval at each site, pediatric patients (≤ 21 years of age) with spinal deformity who underwent primary, revision, and definitive spinal fusion between 2004 and 2018 with standard perioperative care were included. Trained research personnel at each site reviewed patient charts, and a final audit was conducted by the first author to identify discrepancies or ambiguities in the data. Additional chart reviews were requested at each site for clarity to ensure the quality of the data. The unit of analysis was procedures instead of patients as some patients had more than one procedure and these procedures were different in terms of invasiveness.

Preventive care factors included the use of topical vancomycin in the operative site and/or bone graft, povidone-iodine irrigations, multilayered closure, or impermeable dressing, the enrollment quality improvement (QI) programs, such as the Children’s Hospitals’ Solutions for Patient Safety (SPS) program [15] (Appendix 1), and adherence to the institutional perioperative antibiotic prophylaxis guideline (Appendix 2). While, the standard closure technique involves conventional fascial, subcutaneous, and skin closure, multilayer closure incorporates the development of myocutaneous flaps and closure of the deep muscles, typically the paraspinous muscles, in a fashion [16]. The technique helps obliterate the peri-hardware dead space and relives tension. The impermeable dressings, such as medical skin adhesive gluing the edges of an incision closed, were waterproof and impermeable to bacteria and contaminants.

The SPS program was designed to prevent patient harms including healthcare-associated infections (HAIs) by employing cultural transformation strategies focusing on in-depth evaluation and change in communication, team dynamics, and leadership. SPS was founded in 2012 and participating hospital enrollment took place from 2012 to 2014. Data regarding adherence to the institutional perioperative antibiotic prophylaxis guideline were collected but were available only from one participating institution. Adherence to the institutional perioperative antibiotic prophylaxis guideline was categorized into incorrect or correct for administration of preoperative, intraoperative, or postoperative dosing or timing (Appendix 2). Patients with suspected infections before surgery who were continued on antibiotics > 24 h after surgery were not categorized as receiving incorrect perioperative prophylaxis.

Data collectors used the Centers for Disease Control and Prevention (CDC) definition of SSI published in 2017 [17] describing SSI as occurring within 90 days after the procedure and involving the skin or subcutaneous tissue of the incision or the fascial and muscle layers below the subcutaneous layer of the incision. In each setting, the treating physicians and surgeons as well as perioperative care team and pediatric infectious disease consults were responsible for closely monitoring wound cultures and readmissions after spine surgery to diagnose and document the occurrence of SSI. If a patient had more than one procedure before an SSI occurrence, the SSI was attributed to the most recent procedure.

Power analysis

An a priori power analysis was conducted to compare the risk of SSI between patients who received each preventive care strategy and those who did not. Given the sample size (n = 3092) and a significance level of 5%, more than 90% power would be achieved to detect absolute difference of 2% in area under curves (AUC). To ensure adequate power to add a care factor in the previously developed risk calculator using prediction modeling, the event per variable (EPV) ratio, defined as the number of outcomes divided by the number of risk factors in the model, for more than 10 was sought [18, 19]. There were 3092 procedures and 132 SSI (4.5%) in the database. Therefore, the prediction model up to 13 predictors was adequate to be entered in the prediction modeling at a time. There were ten patient, surgical and hospital factors already determined; therefore, it was appropriate to include up to three preventive care factors in the final prediction model.

Statistical analysis

Missing data in preventive care factors (2.2–8.9%) were addressed using multiple imputation since there was no evidence that the missing data were not random [20,21,22] (Appendix 3). Simple logistic regression models were first utilized for each preventive care factor to examine its association with the SSI. Significant preventive care factors in the univariable analyses (p < 0.05) and patient, surgical, and hospital factors identified in the risk calculator as significant were included: overweight/obese, neuromuscular etiology, American Society of Anesthesiologist Physical Status Classification System (ASA) > 2, non-ambulatory status, abnormal hemoglobin level, revision surgery, presence of pelvic instrumentation, procedure time ≥ 7 h, and 100 spine surgical case per year per institution. The corresponding odds ratio (OR) of SSI in the multiple logistic regression model was obtained for each preventive care factor.

The data were randomly split into training (80%) and testing (20%) cohorts, and five-fold cross-validation was performed with model fit conducted only in the training sets [23]. The AUC plotting sensitivity vs 1-specificity were calculated to evaluate the model’s ability to discriminate patients with and without SSI [24]. Discrimination abilities were further assessed by discrimination slopes and box plots comparing average prediction differences in those with and without the observed SSI [25], and by Lorenz estimates and curves depicting the cumulative proportion of patients ranked by predicted probability against the cumulative proportion of patients with SSI [26]. The model calibration was assessed by the Hosmer–Lemeshow (HL) goodness-of-fit test along with the graphic illustration of the fit using calibration plots [27]. Calibration slopes [28], and calibration-in-the-large [29] were also evaluated. Overfitting was calculated by in-sample error over out-sample error comparing the average deviances and Pearson’s residuals of training sets and testing sets. To calculate the individual probability of SSI, a risk prediction algorithm was created from coefficients in the final model. Additionally, a smart phone application for the dynamic calculator was developed to facilitate use in clinical settings.

Results

There were 3092 spinal deformity surgical procedures, and a total of 132 SSI within 90 days after surgery were reported (4.5%) (Table 1). Multilayered closure was performed in approximately half of patients followed by topical vancomycin in approximately 40% of patients, povidone-iodine irrigations and QI enrollment in about one-fourth of patients, and impermeable dressing in approximately 15% of patients (Table 2). For the institution with available data regarding adherence to their perioperative antibiotic prophylaxis guideline, there were 1487 surgical procedures and 57 SSI (3.8%) (Table 3). Reported adherence to postoperative dosing of antibiotic prophylaxis had the lowest reported adherence (78.3%) and postoperative timing had the highest (91.5%). Adherence to preoperative and intraoperative dosing and timing was similar, ranging from 87 to 89.5%. Univariable regressions demonstrated that enrollment in QI programs and povidone-iodine (PI) irrigation was significantly associated, and topical vancomycin, multilayered closure, and correct intraoperative dosing of antibiotics were trended toward association with reduction of SSI (Tables 2, 3). When the SPS enrollment alone was in the model, patients whose procedures were performed when sites were enrolled in the programs had 49.4% decrease in SSI (odds ratio [OR] 0.51, [95% CI 0.32; 0.81], p = 0.005) and AUC of 0.56 [95% CI 0.52; 0.59].

Table 1 Descriptive Statistics for baseline characteristics
Full size table
Table 2 Descriptive statistics for preventive care factors in all patients
Full size table
Table 3 Descriptive statistics for adherence to IV antibiotics in the subgroup
Full size table

The final model using multiple regression including povidone-iodine irrigations and the enrollment in SPS as well as the previously identified patient, surgical, and hospital characteristics demonstrated adequate predictive discrimination and calibration abilities in the training and testing sets (Appendix 4). The average discrimination abilities of this model in the training and the testing sets were AUC: 0.78 [95% CI 0.74; 0.83] and 0.77 [95% CI 0.69; 0.85], the discrimination slope of 0.05 [95% CI 0.04; 0.06] and 0.05 [95% CI 0.03;0.06], and Lorenz curve: 2.81%, 12.87%, and 35.65% and 3.52%, 12.19%, and 39.37% at 25%, 50%, and 75% cumulative risk proportions respectively. The average calibration abilities were calibration slope: 1.03 and 0.97, expected/observed ratio: 0.99 and 0.99, calibration-in-the-large: 0.01 and 0.01, HL goodness-of-fit-tests of 0.002 and 0.16. Overfitting was not observed: deviance of 0.99 and Pearson’s residuals of 0.99. Coefficient and odds ratios with 95% confidence intervals (CIs) for preventive care factors in the final prediction model are presented in Table 4. Patients from institutions enrolled in the SPS programs had an average 48.9% lower in SSI compared with patients from non-enrolled sites [odds ratio: 0.51, (95% CI 0.30; 0.86), p = 0.01]. Although not statistically significant, patients who received povidone-iodine irrigations had an average 18.3% decrease in SSI compared with patients without the irrigations [odds ratio: 0.81, (95% CI 0.44; 1.48), p = 0.494]. Accumulative AUCs are presented in Appendix 5.

Table 4 Odds ratio and coefficients for SSI in the final prediction model
Full size table

The equation and the smartphone application to calculate the reduction of the predicted risk of SSI in individual patients from the final model are presented in Figs. 1 and 2 respectively. Some likely cases are described in Table 5. For example, Case 2 shows the predicted risk of SSI in patients with neuromuscular etiology, ASA = 2, and non-ambulatory status, who underwent pelvic instrumentation and had ≥ 7-h surgical procedure. The predicted risk was reduced from 13.4% to 11.1% with povidone-iodine irrigations, to 7.3% with SPS enrollment, and to 6.0% with both strategies.

Fig. 1
figure 1

Equation calculating the reduction of individual probability of SSI by preventive care factors

Full size image
Fig. 2
figure 2

App for the dynamic risk calculator producing reduction of individual probabilities of SSI by preventive care factor

Full size image
Table 5 Predicted risk of SSI with preventive strategies
Full size table

Discussion

The goal of this study was to identify preventive care factors associated with a reduction in SSI incidence within 90 days of pediatric spinal deformity surgery. The previously developed risk calculator used prediction modeling to identify the probability of SSI in individual patients based on preoperative factors and intraoperative factors, which are determined and planned preoperatively and unlikely to be modified in many cases. In this study, we attempted to identify modifiable preventive care factors in individual patients, taking the baseline risk of these patients into account as calculated by the previously developed clinical risk model [14].

This study demonstrated that enrollment in SPS was most significantly associated with a reduced risk of SSI. SPS, focusing on teamwork, communication, and leadership, was designed to prevent patient harms by facilitating organizational improvements and employing cultural transformation. This finding was consistent with existing literature reporting the benefit of QI programs in reducing SSI in various surgical specialties among both adult and pediatric populations [30,31,32,33]. This suggests that the socio-adaptive aspects of care were especially important in reducing the risk of SSI. Quality improvement requires orchestrated efforts including robust leadership and commitment of the entire care team along with an understanding of health care delivery and human behavior [34, 35]. Hence, reviewing and investing in socio-adaptive aspects of care delivery may be a crucial step along with seeking other technical approaches. Additionally, SPS is a multimodal QI program that may improve the administration and adherence to other preventative strategies examined, such as antibiotic timing.

On the technical side, this study showed that povidone-iodine irrigations might have been associated with a reduced risk of SSI. Although it was not statistically significant, on average, the calculated probability of SSI at the individual level produced by the equation was still reduced when povidone-iodine irrigations were performed. The impact of the povidone-iodine irrigations found in our study was not as great as in a previous study reporting an approximately 20 percent reduction [36]. This may be because our sample size was too small, or the previous study only included patients with adolescent idiopathic scoliosis (AIS) while this study had patients with AIS as well as younger patients and patients with more involved etiologies and complex comorbidities.

This study is important in several ways. First, identifying potentially modifiable preventive care factors is valuable in the clinical setting. Although it is not possible to modify some risk factors, such as neuromuscular etiology, changing modifiable factors can reduce the incidence of SSI. Targeting preventive care strategies for children undergoing surgery for spinal deformities, and focusing on those identified to be at the highest risk for SSI, is important and may enable simultaneous improvements in the quality of care while minimizing per capita costs [13]. Second, results of this study can provide insights into potential causal mechanisms of SSI. Although risk prediction and investigation of causal inference differ in principle and methodology, prediction modeling identifies exposures significantly associated with outcomes. Therefore, preventive care factors identified in this research warrant further study to advance our understanding of potential strategies to reduce SSI. Third, this study defined SSI using the standard CDC definition which is also used by Center for Medicare and Medicaid Services to determine penalties for surgical readmission.

There were several limitations in this study. First, misdiagnosis or variations in defining SSI might have occurred across sites as SSI ranged from 0 to 5.3%. Although each site agreed to use the CDC definition of SSI and was asked to validate the SSI data in patient charts, the accuracy of the data was dependent on research personnel at each site. Despite providing each site with a standardized definition, variations in reporting may also have occurred for multilevel closure due to differences in surgical technique. Second, an SSI could have been treated at an outside hospital not contributing to the database and not reported to a performing surgeon. However, this was unlikely due to the seriousness of spinal surgery and SSI. Next, some preventive care factors may have been misclassified or not recorded. If recording errors of the outcome and/or exposures were different, information bias and inaccurate prediction are possible. An important next step in the research is to validate the reduction effect of preventive care factors tested in this study in multiple data sets across different times and settings. Finally, the SPS program is specific to the United States and may not be generalizable to other countries. SPS focuses on the socio-adaptive aspects of care management, facilitated organizational improvements and cultural transformation, but may not be appropriate in different cultures and customs. Therefore, future studies which investigate causal pathways (mediator effects) between quality programs and the decreased risk of SSI are needed to identify potential interventions to replace culturally specific programs that improve human behavior and reduce the risk of SSI in other countries. Additionally, future studies could investigate how the level of site involvement in QI programs interacts with SSI risk.

Conclusion

In conclusion, this study presents the first-time evaluation of the potential effects of preventive care factors on SSI risk in individual patients, considering individual patients’ baseline characteristics and predetermined surgical and hospital factors, which are difficult to modify and can confound the results. The final model encompassing preventive care factors and patient, surgical and hospital factors has adequate predictive accuracy for 90-day SSI after surgery in pediatric patients with spinal deformity. SSI incidence was most significantly associated with quality improvement program enrollment, further supporting the use of multimodal, multidisciplinary teams to improve patient safety. The results of this study add new information to enhance personalized care in clinical practice by identifying factors, which could reduce the risk of SSI for specific patients.