Clinical Trial Details
— Status: Withdrawn
Administrative data
| NCT number |
NCT03982810 |
| Other study ID # |
2000023706 |
| Secondary ID |
|
| Status |
Withdrawn |
| Phase |
|
| First received |
|
| Last updated |
|
| Start date |
June 1, 2019 |
| Est. completion date |
March 2024 |
Study information
| Verified date |
July 2022 |
| Source |
Yale University |
| Contact |
n/a |
| Is FDA regulated |
No |
| Health authority |
|
| Study type |
Observational
|
Clinical Trial Summary
This study will seek to describe current practice of antibiotic prophylaxis to identify the
effect of appropriate perioperative antimicrobial coverage - specifically regarding timing,
dose adjustments, and redosing - on surgical site infections (SSI).
Description:
Introduction:
Prevention of surgical site infection (SSI) continues to be a major challenge for the health
care system since it incurs a substantial toll on public health and significantly inflates
health care costs. SSIs are now the leading cause of health care related infections,
complicating about 2-5 % of all surgeries(1-3). SSIs affects about 125,000 cases annually
accounting for nearly a million excess hospital days and just under $1.6 billion in
additional health care costs(4). It is estimated that half of the SSIs are preventable(5) and
not surprisingly, the prevention of health care-associated infections has been a priority
objective of the U.S. Department of Health and Human Services (HHS)(6) over the past several
years. Public reporting of SSI outcomes is now mandatory and reimbursement for management of
SSIs is being reduced or denied(7,8) in an effort to curb its incidence.
Despite the institution of stringent measures and surveillance programs, surgical registries
continue to show SSI rates of about 2-5%(9,10) and SSIs remain a key cause of prolonged
hospitalization, morbidity and death. The continued health care burden caused by SSI calls
for closer scrutiny of the current clinical practices especially pertaining to perioperative
antibiotic coverage. Although the institution of timely perioperative antibiotic prophylaxis
is now a National Quality Anesthesia Care Measure(11), much remains to be known about
antibiotic redosing, weight based adjustments and completion of antibiotic infusion prior to
skin incision(7).
This study will seek to describe current practice of perioperative antibiotic prophylaxis
among MPOG institutions, and in the subset of MPOG centers contributing NSQIP data, to
identify the effect of appropriate guideline-based perioperative antimicrobial coverage -
specifically regarding selection, timing, weight-based dose adjustments, and redosing - on
SSI. The applicable guidelines against which adherence will be assessed are those of the
Infectious Disease Society of American (IDSA). To understand the effects of IDSA guideline
adherence, the investigators propose to utilize the American College of Surgeons - National
Surgical Quality Improvement Program (ACS-NSQIP) data collection methodology, and to
integrate these prospectively collected outcome data from 6 centers within MPOG with
intraoperative anesthesia electronic health record (EHR) data available within MPOG. Beyond
the descriptive aim to describe current practice, our primary inferential hypothesis is that
adherence to IDSA guidelines regarding appropriate antibiotic selection, timely antibiotic
redosing, weight based dose adjustments, and appropriate timing of infusions to ensure
completion of administration prior to skin incision will be associated with a lower incidence
of SSIs when considered both individually and as a basket of practices, while controlling for
common confounders available within the MPOG and NSQIP datasets.
Methods Approval has been obtained from the Yale IRB for this multicenter, observational
retrospective study. Data have previously been collected under an umbrella IRB protocol
within the University of Michigan. The ACS-NSQIP methodology has been described in detail
elsewhere(12). For the NSQIP/MPOG portion of the study, data collected from 01/01/2011 to
07/04/2018 will be extracted. Since the IDSA guidelines were proposed in 02/2013, for the
descriptive portion of the study looking at predictors of guideline adherence, data from
01/01/2014 to 07/04/2018 will be extracted from the MPOG database.
Patient population All patients equal or greater than 18 years of age undergoing non-emergent
non-cardiac surgical procedures involving a skin incision will potentially be included in the
study. For the NSQIP/MPOG portion of the study, patients with conditions that could confound
the analysis of SSI risk factors including emergency surgery, open wound with or without
infection, current active infection, ongoing preoperative antibiotic therapy, missing
perioperative antibiotic/medication documentation, ventilator dependence within 48 hours of
surgery, ophthalmic surgeries, organ transplants, prior operation within 30 days, Organ
harvesting surgeries, and ASA 5 or 6; will be excluded. A complete list of the exclusion
criteria from ACS-NSQIP variables is documented in Supplement 1. For the descriptive study of
MPOG antibiotic practices, exclusions are listed in Supplement 2.
Exploratory Factors:
The following MPOG and ACS-NSQIP preoperative clinical variables will be evaluated for its
relationship with the occurrence of SSI in the primary inferential analyses (parentheses
indicate the source database): age (MPOG), male sex (MPOG), body mass index (MPOG), diabetes
mellitus (NSQIP, current smoker within 1 year (NSQIP), severe COPD (NSQIP), congestive heart
failure within 30 days (NSQIP), history of myocardial infarction (NSQIP), hypertension
(NSQIP), history of peripheral vascular disease (MPOG), ongoing dialysis requirements
(NSQIP), transient ischemic attacks or stroke (NSQIP), disseminated cancer (NSQIP), loss of
10% of body weight in 6 months (NSQIP), steroid use for a chronic condition (NSQIP),
chemotherapy within 30 days (NSQIP), and ASA physical status (MPOG).
Body mass index will be transformed into categorical variables based upon the clinically
relevant World Health Organization classification scheme (< 20, 20-25, 25-30, 30-35, 35-40,
40-50, and > 50 kg/m2). ASA physical status will be transformed into three categorical dummy
variables: ASA 1, 2, 3 or 4. Diabetes mellitus will be transformed into two dummy variables:
diabetes mellitus requiring oral hypoglycemic treatment without insulin (NSQIP), and diabetes
mellitus requiring insulin treatment with or without oral hypoglycemic (NSQIP).
Intraoperative variables including hypotension, hypothermia, transfusion volume, the need for
vasopressor / inotrope infusion, median fiO2 and surgery duration will be included.
For intraoperative variables, hypotension will be calculated as the time in minutes below MAP
55mmHg. Transfusion volume will be calculated as the number of pRBC units transfused between
surgery start and surgery end. The need for infusions of vasopressors and/or inotropes will
be coded as yes/no based on the intraoperative anesthetic record and including only the need
for infusions without regard for isolated bolus dosing. Duration of surgery will be
calculated as the period of time from incision to surgery end. Median FiO2 utilized during
the surgeries will be calculated.
Although there are a number of studies reporting the effect of hypothermia on SSI after
certain surgeries, a consensus on a metric to measure the magnitude of hypothermia associated
with SSI is lacking. In addition, intraoperative temperature measurement is subject to
numerous artifacts such as dislodgment of the temperature measuring device from the patient.
To minimize this, the investigators will utilize the artifact reducing algorithm. After
artifact removal, the median temperature will be calculated for use in the relevant models.
Endpoints:
The primary end point to which the investigators will attempt to associate guideline-adherent
antibiotic prophylaxis will be occurrence of a NSQIP-adjudicated SSI during the period from
01/01/2011 to 07/04/2018. SSIs will be a composite of superficial (only skin or subcutaneous
tissue of the incision), deep (deep soft tissues), and organ space (any part of the anatomy
other than the incision, which has been opened and manipulated during the operation), as
provided by NSQIP.
Appropriate antibiotic prophylaxis:
Definition for appropriate antibiotic prophylaxis will be used per the Infectious Diseases
Society of America (IDSA), the Surgical Infection Society (SIS) and American Society of
Health-System Pharmacists (ASHP) guidelines(13). Data on timing, dose, redosing and choice of
antibiotics will be obtained from MPOG.
Choice of antibiotics:
The IDSA guidelines will be utilized to assess choice of antibiotics (Supplement 3).
Appropriate antibiotics will be decided a priori for all the CPT codes based on these
guidelines. Patients will then be classified into 2 groups based on the "choice of
antibiotics." Under certain patient/hospital-based scenarios, the guidelines recommend
additional antibiotics or a preference towards a certain class among the listed antibiotics
in the category. The plan is to consider the antibiotic choice as appropriate if any
antibiotic from the listed procedural category is utilized for the surgery. In case more than
one antibiotic is administered, at least one antibiotic or a combination of antibiotics
should match the recommendations.
Timing of antibiotics with respect to surgical incision will be coded in two ways, first as a
continuous variable to assess the nonlinear association between antibiotic timing and
SSI(14). Second, timing of antibiotics will be dichotomized whether it fits in the time
period of existing guidelines and assessed as a categorical variable for its association with
SSI. For antibiotic infusions, the start of antibiotic will be considered as time of
administration.
Dosing with respect to weight adjustment will be considered in reference to the same
guidelines and considered adherent if the dose of the appropriate antibiotics meets the
minimum requirement for weight-based adjustment. For antibiotics with weight based guidelines
in mg/kg (example vancomycin), dosing up to 10% below the calculated dose will be considered
as guideline adherent.
Redosing will be considered in a dichotomous fashion and will be coded as adherent if the
surgical duration necessitated a guideline-indicated redosing interval and such a dose was
administered prior to that interval. In cases in which more than one redosing episode should
have occurred, redosing adherence will be considered in an all-or nothing fashion whereby a
lack of any timely guideline-adherent redosing will be coded as non-adherent.
Trends in guideline-adherent antibiotic usage:
The investigators will also investigate the trends in guideline-adherent antibiotic practices
within the MPOG database including those institutions not contributing NSQIP data as per
exclusions in supplement 2. This analysis will consider within-institution temporal trends
and will examine the possible association of candidate patient-level and institution-level
factors. More specifically, the rates of guideline-adherent antibiotic practices will be
modelled using the mixed-effects multiple logistic regression method that include fix effects
such as time (and polynomial terms of it if non-linearity is confirmed), institution-level
factors (e.g., institution type, size, etc.) and patient-level variables, and random
institution effect. The significance (i.e., p < 0.05) of coefficient for the time variable
will be indicative of a significant overall trend effect. The estimates of adherence rate and
their 95% confidence intervals (CIs) will be calculated.
Statistical analysis:
Statistical analysis will be performed using SAS version 9.4 (Cary, NC). A two-sided p-value
<0.05 will be considered statistically significant, if not otherwise noted. Appropriate
effect sizes (e.g., odds ratio), and their corresponding 95% confidences intervals (CIs) will
be reported. Descriptive statistics (means, medians, frequencies) will be calculated to
characterize demographics and all extracted clinical variables. Histograms and box plots will
be constructed to evaluate distributions of continuous variables and identify potential
outliers. Each outlier will be reviewed carefully and verified. Categorical items with more
than two categories that do not exhibit sufficient variability across response levels will be
dichotomized accordingly.
For the descriptive aim in parallel with the above analysis, practice patterns across MPOG
institutions in relation to antibiotic selection, dosing, redosing, and timing will be
examined. The distribution of adherence to these practices will be examined, and patient,
provider, and institution level predictors of adherence to these practices, individually and
as a bundle will be examined. Box-plots, caterpillar plots, and funnel plots will be
generated to visualize the patterns/variability of SSI rates and potentially point out
unusual performers at both local (i.e. institution) and national levels. In a typical funnel
plot, the institution-specific rates can be plotted against the institution case volume with
95% and 99% confidence limits (corresponding to 2 and 3 standard deviations) superimposed
around the rates. Institutions and providers with rates out of these limits will be marked as
"outliers" and subject to further scrutinization to under the reason for the abnormal
variability.
For the primary inferential aim, univariate analyses will be first performed using Pearson
Chi-Square, Fisher's Exact Test, Student's t-test, and Mann Whitney U Test as appropriate to
investigate the association of all preoperative and intraoperative variables with the outcome
of NSQIP-adjudicated SSI. Generally, only the factors with p 0.1 from univariate analysis
will be included in the multivariable regression model. However, Clinical variables with
shown evidences affecting the risk of SSI will also be included in the model. Collinearity,
the linear assumption, and the additivity assumption of the predictors will be checked, and
nonlinear modeling of continuous predictors (e.g., infusion time) will be investigated. If
necessary, highly correlated groups of predictors will be examined and dimensionality will be
reduced either by subject matter knowledge (i.e., principal components), or by simple point
scores.
After examining the prevalence or patterns of SSI by different center or surgery types, four
distinct clustered or mixed-effects multiple logistic regression models to will be developed
using SAS GLIMMIX procedure to associate the SSI outcome with each component of
intraoperative antibiotic management domains: choice, redosing interval, weight-based
adjustment, and time of administration criteria. Specifically, the investigators propose to
test the hypothesis that correct antibiotic choice, timely antibiotic dosing, redosing,
weight-based dose adjustments in accordance with guidelines, appropriate timing of infusions
to ensure completion of administration prior to skin incision will be independently
associated with a lower incidence of SSIs while controlling for significant confounders.
Random effects for hospitals and anesthesia providers will be included to address the
clustering of different surgical cases. The investigators will examine the modification
effects of other specific factors, adding them into the model as fixed factors, which include
patient level demographics such as age, health of patient (ASA class), BMI, gender,
race/ethnicity, and ACS-NSQIP preoperative and other operative variables. In addition to
p-values, as the measures of effect sizes, the investigators will also report adjusted odds
ratios and 95% confidence intervals for each independent variable in the final model,
comparing the likelihood of SSI among patients with and without the risk factor.
A dummy variable will be created that is coded as 'Yes' if adherence to guidance for all four
intraoperative antibiotic management domains are met or 'No' otherwise, then the association
of this dummy variable with the likelihood of SSI will be tested. This would help quantitate
a composite effect for adherence to guidance on the SSI. Finally, an overall model
incorporating all domains, preoperative and operative ACS-NSQIP variables, and the surgical
complexity score will be developed using the same methodology described above.
For the purpose of model performance diagnosis, the amount of variability in the SSI outcome
that is explained by each regression model will be quantified by the adjusted-R2 statistic,
and the discrimination performance of the model will be assessed by C-statistic (i.e. AUC).
The Hosmer-Lemeshow goodness-of-fit (GOF) test will be used to check if the final model fits
the data well. A GOF P-value > 0.05 will generally indicate whether a model is a good fit or
well-calibrated. The model will be internally validated using a resampling bootstrap
technique to assess for the possibility of overfitting.
It is worth noting that there were approximately 9.3 million unique cases in MPOG as of June
2018 (and growing monthly), with adequate numbers of patients to develop a descriptive
regression model with a number of variables. Given the rule of thumb of maintaining at 10
events per variable (EPV) in the multivariable logistic regression model, the investigators
will have more than sufficient numbers to precisely estimate up to hundreds of predictors
(when applicable, different categories for a discrete variable is counted as a predictor) in
the final model. That is, overfitting will likely not a concern in the current study.
However, the investigators will closely evaluate the issue when developing our models. If EPV
>= 10 can't not seem to be guaranteed, the investigators will choose to use the penalized
method-the least absolute shrinkage and selection operator (LASSO) for variable (feature)
selection to first create a subset of potential important predictors, which then will be
subject to our standard variable selection procedure described above to select the final
specification of list.
Power analysis:
Although this is an observational analysis that does not involve recruitment of patients, a
power analysis to establish that the database can detect a clinically meaningful and
statistically significant difference is important. Previous SSI prevention interventions such
as normothermia, antibiotic prophylaxis, and chlorhexidine surgical prep have demonstrated
relative risk reduction rates ranging from 40% to 70%. For purposes of this power analysis,
it will be assumed that a conservative benefit of only 20% for each of the intraoperative
interventions, or the group as a "bundle." Review of literature demonstrates a composite SSI
incidence of about 4%. A 20% relative reduction would result in an observed SSI rate of 3.2%.
Assuming the rate of "appropriate antibiotic usage" is 92%, a chi square test with a 0.05
two-sided significance level will have 80% power to detect the difference between these two
rates when a total sample size is 55,637. In aggregate, the institutions presented in this
proposal already offer sufficient ACS-NSQIP cases with integrated anesthesia EHR data.
Prespecified sensitivity analyses:
1. A sensitivity analysis will be conducted in which an attempt to create a
propensity-score matched cohort of patients receiving vs. not receiving guideline
adherent antibiotic prophylaxis to measure the possible association of such adherence to
the same SSI outcome as above.
In this sensitivity analysis, instead of regression covariate adjustment in our primary
analysis, the investigators will use the propensity score method for covariate
adjustment of potential confounding. The propensity scores will be developed to predict
those receiving vs. not receiving guideline adherent antibiotic prophylaxis to address
potential issues of selection bias. Propensity scores will be developed using logistic
regression models to predict exposure group using a dichotomous outcome indicator
variable for exposure (1= receiving guideline-based antibiotics, 0 = not receiving
guideline based antibiotics). The investigators will select a non-parsimonious set of
covariates as listed in the primary analytic modelling description. Then, patients in
two exposure groups will be matched, first via exact matching by institution, patient
age in years, and anesthesia CPT code, followed by propensity score matching using the
greedy method implemented in the %GMATCH SAS macro (Mayo Clinic, Rochester,
Minnesota),or a similar algorithm based on the proximity of individual propensity
scores. To assess whether appropriate balance on covariates has been achieved between
each grouping, standardized difference (d) for each covariate will be calculated. If
this meets a threshold value < 10%(15) the covariates will be considered to be generally
well balanced. If residual imbalance exists and is deemed significant, iterative
recalculation of propensity scores with additional candidate covariates will be
considered. Last, simpler mixed-effects multiple logistic regression models of the SSI
outcome with the fixed effect of exposure variable will be fit. If exact matching within
institution causes diminution of successfully matched samples, the investigators will
consider removing it from the exact match and including a random institution effect
within the final propensity-score matched analysis.
2. Considering the issue of diminution of sample size during matching, another sensitivity
analysis will be conducted in which the method of the inverse probability of treatment
(exposure) weighting using derived propensity scores will be used to compare the SSI
outcome between two groups (1= receiving guideline-adherent antibiotic prophylaxis, 0 =
not receiving), as the weighting method will enable us to include all patients into the
final analysis.
3. In the case of a significant association between guideline adherent antibiotic
administration and SSI, we will explore how prevalent and powerful an unmeasured
confounder would have needed to be able to erase the observed difference. That is, model
the robustness of an observed association in the face of a hypothetical unmeasured
confounder as described by Lin, et al(16). For this analysis a model of the
characteristics of a hypothetical unmeasured binary confounder that could have accounted
for observed differences in odds of SSI between patients with adherent vs. non-adherent
antibiotic dosing, using a broad range of plausible values for the effect size and
prevalence of such an unmeasured confounder.
4. As mentioned above, for antibiotics with weight-based guidelines in mg/kg (example
vancomycin), dosing within 10% of the calculated dose will be considered as guideline
adherent. To assess the correlation of dosing of these antibiotics on SSI, also a
sensitivity analysis will be performed by categorizing patients within 25% of the
calculated dose in guideline adherent group.
