Non-compliance with clinical guidelines increases the risk of complications after primary total hip and knee joint replacement surgery

Background Total hip and total knee replacement (THR/TKR) are common and effective surgeries to reduce the pain and disability associated with arthritis but are associated with small but significant risks of preventable complications such as surgical site infection (SSI) and venous-thrombo-embolism (VTE). This study aims to determine the degree to which hospital care was compliant with clinical guidelines for the prevention of SSI and VTE after THR/TKR; and whether non-compliant prophylaxis is associated with increased risk of complications. Methods and findings A prospective multi-centre cohort study was undertaken in consenting adults with osteoarthritis undergoing elective primary TKR/THR at one of 19 high-volume Australian public or private hospitals. Data were collected prior to surgery and for one-year post-surgery. Four adjusted logistic regression analyses were undertaken to explore associations between binary non-compliance and the risk of surgical complications: (1) composite (simultaneous) non-compliance with both (VTE and antibiotic) guidelines and composite complications [all-cause mortality, VTE, readmission/reoperation for joint-related reasons (one-year) and non-joint-related reasons (35-days)], (2) VTE non-compliance and VTE outcomes, (3) antibiotic non-compliance and any SSI, and (4) antibiotic non-compliance and deep SSI. Data were analysed for 1875 participants. Guideline non-compliance rates were high: 65% (VTE), 87% (antibiotics) and 95% (composite guideline). Composite non-compliance was not associated with composite complication (12.8% vs 8.3%, adjusted odds ratio [AOR] = 1.41, 95%CI 0.68–3.45, p = 0.40). Non-compliance with VTE guidelines was associated with VTE outcomes (5% vs 2.4%, AOR = 2.83, 95%CI 1.59–5.28,p < 0.001). Non-compliance with antibiotic guidelines was associated with any SSI (14.8% vs 6.1%, AOR = 1.98, 95%CI 1.17–3.62,p = 0.02) but not deep infection (3.7% vs 1.2%,AOR = 2.39, 95%CI 0.85–10.00, p = 0.15). Conclusions We found high rates of clinical variation and statistically significant associations between non-compliance with VTE and antibiotic guidelines and increased risk of VTE and SSI, respectively. Complications after THR/TKR surgery may be decreased by improving compliance with clinical guidelines.

This letter describes our response to the issues identified in the initial review. I have attached annotated copies of the initial manuscript submitted to PloS One (2018) and the subsequent manuscripts submitted to PLOS Medicine and transferred to PLOS One (2021) to highlight the changes compared to the current submission. We have also made some further revisions to fully address feedback from the previous review, and a tracked changes and clean version of the manuscript we would hope undergoes further review are also attached. As requested, I have provided responses in relation to the specific questions and issues raised: 1. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer: No We believe that the revised manuscript describes a technically sound piece of scientific research with data that supports the conclusions. Based on the feedback provided by the PLOS ONE reviewers we have completed new analyses of the data and adjusted the statistical methods to increase the rigour of the data analyses.
In our initial submission, our conclusion was that there was no significant impact of non-compliance on overall composite outcome. The magnitude of effect for the primary outcome comprising composite surgical complications was 13% in the non-compliant group versus 8% in the group that received care considered compliant with both the guidelines for the prevention of VTE and antibiotic prophylaxis. This difference did not reach statistical significance. In this iteration we have also included new analyses examining the association between VTE compliance and VTE outcomes, and antibiotic compliance and infection outcomes, both of which were statistically significant. We believe these additional analyses strengthen the scientific merit of the manuscript and clinical relevance of the results.
The revised methods now include: "We conducted multiple logistic regression analyses to explore associations between non-compliance and risk of outcomes as follows: (1) non-compliance with both VTE and antibiotic guidelines and composite outcomes (2) VTE non-compliance and VTE outcomes, (3) antibiotic compliance with any SSI and (4) antibiotic compliance with deep SSI. Factors identified on univariate analysis with a p-value < 0.25 were entered into a backwards, stepwise multivariable logistic regression model (using the Akaike information criterion -AIC and forcing the main predictor, compliance, into the model) to identify independent risk factors associated with the relevant outcome for each analysis.
Missing data were imputed using multivariate imputation by chained equations (MICE). Model selection was performed using one of the imputed datasets. Effect estimates were taken from the pooled estimates using the five imputed datasets.
Sensitivity analyses were performed using complete case analysis and Bayesian information criterion (BIC). Further sensitivity analyses were completed without including routine doppler ultrasound (DUS) given this may mediate VTE complication outcomes. The full R code for all analyses are included in a supplementary file (Supplementary File 2)."

Has the statistical analysis been performed appropriately and rigorously? Reviewer: No
We have completed new and more extensive analyses in light of the reviewers' feedback as outlined in the methods above. The action taken for each suggestion are reported in more detail in our responses to the reviewers specific comments further below under General Comments.

Have the authors made all data underlying the findings in their manuscript fully available?
We initially indicated that data from this study are available upon request. We have clarified approval to make a deidentified data set including individual patient level data publicly available with the Lead Human Research Ethics Committee. We have uploaded this deidentified dataset to UNSW Data Archive and it will be published and publicly available via Research Data Australia https://researchdata.edu.au/ . The DOI for the data is still pending and was due to be confirmed on 18/03/2021. We can confirm this once known.
We believe the provision of this data is consistent with the PLOS Data policy, and should enable readers and reviewers to validate any of the results provided. Thank you for updating the Data Availability statement on our behalf to reflect the revised information that we have provided.

Is the manuscript presented in an intelligible fashion and written in standard English? Yes.
We have edited the revised manuscript to ensure a high standard of academic writing, appropriate use of grammar, and adherence to the required refencing style.

General comments:
Reviewer #1: The authors present the results of an observational cohort study in around 1875 patients with total hip and total knee replacement (THR, TKR) to investigate the compliance with clinical guidelines for prevention of complications. The primary endpoint was a composite comprising allcause mortality, any VTE and any reoperation or readmissions within 35 days for medical issues or within 365 days for joint related complications. The study shows higher rate of complications in the non-compliant group.
I first was very confused by the abstract, where no clear specification of the study type is given. Thus the sampling as well as the potential conclusions could not be followed. To my understanding in the paper the authors conducted a prospective multicentric cohort study to investigate the compliance to clinical guidance for VTE as well as Antibiotic prophylaxis on the composite primary endpoint.

Revision to Abstract
The abstract has been revised to provide clearer information about the type of study conducted. This has included revision of the methods used to improve understanding of the sampling and potential conclusions. The reviewer correctly identified that this was a prospective multi-centre cohort study of people undergoing primary THR or TKR that aimed to determine the degree to which current practice in Australian hospitals was compliant with Australian guidelines to prevent VTE and infection after TKR or THR; and whether non-compliance with these guidelines is associated with increased risk of complications. The current iteration also has the results of the additional analyses.
The abstract now reports:

"Methods and Findings
A prospective multi-centre cohort study was undertaken in consenting adults with osteoarthritis undergoing primary TKR/THR at one of 19 high-volume Australian public or private hospitals." 1. Individual data should be made available according to PLOS philosophy. As reported above a deidentified dataset has been uploaded to the UNSW Data Archive and it will be published and publicly available via Research Data Australia https://researchdata.edu.au/ . The DOI for the data is still pending and was due to be confirmed on 18/03/2021. We can confirm this once known.

Taking into account the sample size argumentation, the general setting of the trial assumes an unadjusted effect. To me it is not clear, whether this is reasonable.
In the initial submission we accounted for the impact of confounders in both the sample size calculations and in our analyses: We used the above reference to estimate the expected levels of compliance for the a priori sample size calculations . Our expected event rate of 12.7% in the non-compliant group compared to 8.3% in the compliant group was close to the values used in our a-priori calculation (7% event rate in the compliant group). However, our compliance rate was very different. We anticipated a 67% compliance, yet reported only 4.5% compliance with both guidelines, 13.2% with the antibiotic guideline and 35.3% with the VTE guideline.
Section Sample Size: Please specify the test procedure used for sample size calculation. Give references to the assumed RR effect and the software used. The sample size calculation is based on a pooled analysis in particular without adjustment for effect modifiers.
The a priori power calculations were performed using the 'procpower' calculation in SAS software for comparing two independent proportions. We used the reference noted above (Bozic et al., 2010) to the manuscript as the source used to estimate the potential prevalence of composite compliance and complications in our data set.

A adjusted analysis should be conducted using first main effects in the analysis. Hereby a suitable bivariate variable selection technique should be used, e.g. by fitting bivariate logistic regression models to the data and decide for variable inclusion by p-values below 25%. Interaction effects should be investigated as well. Results of this investigation have to be presented.
In this current iteration we performed separate unadjusted analyses for the outcomes of each analysis: 1. Composite non-compliance and composite surgical complications.

VTE non-compliance and VTE complications.
3. Antibiotic non-compliance and surgical site infection (SSI). 3.1 Any SSI (requiring oral or IV antibiotics, hospital readmission or reoperation) 3.2 Deep SSI only (requiring IV antibiotics, hospital readmission or reoperation)

"We conducted multiple logistic regression analyses to explore associations between non-compliance and risk of outcomes as follows: (1) non-compliance with both VTE and antibiotic guidelines and composite outcomes (2) VTE non-compliance and VTE outcomes, (3) antibiotic compliance with any SSI and (4) antibiotic compliance with deep SSI. Factors identified on univariate analysis with a pvalue < 0.25 were entered into a backwards, stepwise multivariable logistic regression model (using the Akaike information criterion -AIC) to identify the association between guidelines compliance and complication outcomes for each analysis. We reported the final model after backwards stepwise regression using AIC and forcing only the main predictor (non-compliance) into each model."
In the initial submission we had performed a bivariate logistic regression although used a theoretical rather than an analytic approach as the reviewer has recommended. There are many possible ways to select possible confounders and in the initial submission we used clinical expertise of the authors and the literature to identify the confounders we included in the model. In this current iteration we have used a bivariate variable selection technique by fitting bivariate logistic regression models to the data and decide for variable inclusion by p-values below 25%, for each of the models .
The results of this analyses can be obtained from the code, and we have provided the eligible covariates included in the three models based on this analytical method (prior to backward stepwise regression).