Surgical Approach and Long-Term Recurrence After Ventral Hernia Repair

IMPORTANCE The prevalence of robotic-assisted anterior abdominal wall (ventral) hernia repair has increased dramatically in recent years, despite conflicting evidence of patient benefit. Whether long-term hernia recurrence rates following robotic-assisted repairs are lower than rates following more established laparoscopic or open approaches remains unclear. OBJECTIVE To evaluate the association between robotic-assisted, laparoscopic


R
obotic-assisted surgery is rapidly expanding into new clinical domains. 1 This is especially true for anterior abdominal wall (ventral) hernia repair, where use of robotic-assisted repair increased 45-fold between 2012 and 2018 alone. 2,3When compared with an open procedure, proponents of robotic-assisted hernia repair cite the benefits of a minimally invasive approach, such as smaller incisions and fewer postoperative complications.][6][7] Despite the rapid adoption of robotic-assisted ventral hernia repair, it remains unclear whether long-term outcomes, such as hernia recurrence, are superior for a robotic-assisted approach compared with more established laparoscopic or open approaches.Existing observational studies and randomized trials comparing robotic-assisted, laparoscopic, and open ventral hernia repair are limited in several important ways.7][18][19] Moreover, existing population-level studies demonstrate conflicting data on recurrence after robotic-assisted repair and do not account for selection bias, limiting researchers' ability to infer causality between operative approach and recurrence. 20,21][18] We sought to evaluate long-term recurrence following robotic-assisted, laparoscopic, and open ventral hernia repair among a national cohort of Medicare beneficiaries.This allowed us to evaluate real-world outcomes using populationlevel data with longitudinal follow-up, due to near-universal enrollment and low disenrollment rates.Using Medicare data also allowed us to leverage regional differences in the adoption of robotic-assisted ventral hernia repair as an instrumental variable (IV) to account for confounding from unmeasured factors that may bias comparisons.A better understanding of the comparative outcomes for each approach can inform the evidence-based adoption of roboticassisted ventral hernia repair and further our knowledge of the true benefits of this technology.

Data Source and Patient Population
We used data from the Medicare Provider Analysis and Review Parts A and B files to identify initial (index) operations for patients 18 years or older who underwent elective, inpatient ventral hernia repair from January 1, 2010, through December 31, 2020.Ventral hernia included all anterior abdominal wall hernias coded in claims as ventral, incisional, umbilical, or epigastric hernia repair.Patients were identified initially using International Classification of Diseases, Ninth Revision (ICD-9) and ICD-10 procedure codes, which were crossreferenced with the corresponding ICD-9 and ICD-10 diagnosis codes (eTable 1 in Supplement 1).Patients were excluded from the index cohort if they had a prior hernia repair in the 2 years leading up to the initial operation or if the operation was associated with Current Procedural Terminology (CPT) codes for repair of a recurrent ventral hernia (codes 49565, 49566, 49656, or 49657).All subsequent admissions for a ventral hernia operation after the index repair were excluded from the index cohort.Patients not enrolled in fee-for-service Medicare were also excluded due to lack of accurate follow-up.To ensure we accurately captured index hernia repairs from ICD-9 and ICD-10 procedure codes, patients were also excluded if they lacked a concurrent CPT code for hernia repair (eTable 1 in Supplement 1).A flow diagram with detailed stepwise patient inclusion and exclusion is available in eFigure 1 in Supplement 1.This study used deidentified patient data and was therefore deemed exempt from regulation by the University of Michigan Institutional Review Board.This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Outcome Measures and Explanatory Variables
The primary outcome was operative recurrence for ventral hernia repair.Operative recurrence was used as a proxy for true clinical recurrence, which cannot be measured in Medicare claims data alone.The total number of ventral/incisional and umbilical hernia repairs was tabulated for each year studied.Surgical approach (robotic-assisted, laparoscopic, or open) was identified using corresponding CPT, ICD-9, and ICD-10 codes (eTable 1 in Supplement 1).Operation for recurrence was identified by a subsequent hernia repair using the same CPT, ICD-9, and ICD-10 codes used to identify the initial hernia repair, the presence of specific hernia recurrence CPT codes (codes 49565, 49566, 49656, and 49657), or both methods.
Explanatory variables included in our models were consistent with prior published work on hernia and included patient age, sex, race and ethnicity (eMethods 1 in Supplement 1), Elixhauser comorbidities, year of surgery, approach

Key Points
Question How do long-term operative recurrence rates for anterior abdominal wall (ventral) hernia compare after robotic-assisted, laparoscopic, and open hernia repairs?Findings This cohort study of Medicare claims data identified 161 415 patients who underwent ventral hernia repair from January 2010 to December 2020.Up to 10 years after surgery, patients who underwent robotic-assisted hernia repair had higher rates of operative recurrence (13.4%) than patients who underwent laparoscopic (12.3%) or open (12.7%) repair.
Meaning Higher long-term recurrence rates after robotic-assisted ventral hernia repair compared with other methods call into question the rapid and widespread adoption of this approach over the last decade.

IV Analysis
The rapid uptake of robotic-assisted hernia repair necessitates the use of existing observational data, as opposed to launching a large-scale randomized clinical trial, to assess longterm outcomes after robotic-assisted, laparoscopic, and open ventral hernia repair.However, due to measured and unmeasured confounding in observational data, causal inference may be limited by selection bias.Standard multivariable logistic regression analysis would not account for this selection bias, as it can only adjust for measured covariates.One known strategy to mitigate this bias is to perform an IV analysis.For additional information, including the reasoning behind and justification for the choice of an IV analysis, please see eMethods 2 in Supplement 1.
The IV used for this analysis was the rate of use of roboticassisted ventral hernia repair within a hospital referral region (HRR) in the 12 months prior to a patient's initial ventral hernia repair.Using prior-year state-level and regional-level procedural use as an IV is consistent with previous observational studies in surgical patients. 22,23We excluded patients located in HRRs with no robotic-assisted hernia repairs during the study period.

Statistical Analysis
Due to the binary, nonlinear outcome of operative hernia recurrence, time-to-event analysis was performed using a 2-stage residual inclusion estimation method for our IV model. 24For the first stage, multivariable logistic regression was used to estimate the likelihood that a patient would undergo roboticassisted ventral hernia repair, while adjusting for the following covariates: prior-year HRR use of robotic-assisted ventral hernia repair (the IV), age, sex, race and ethnicity, comorbidities, myofascial flap use, and hernia subtype.Mesh use was excluded from this first stage, as it was perfectly collinear with robotic-assisted use.For the second stage, a Cox proportional hazards model was constructed to calculate hazard ratios (HRs) and the cumulative incidence of operative hernia recurrence, while adjusting for the following covariates: surgical approach, age, sex, race and ethnicity, comorbidities, hernia subtype, myofascial flap use, mesh use, year of surgery, and residuals from the first-stage regression model, which represent unmeasured confounding associated with the choice of operative approach.The second stage of this model also accounted for clustering of outcomes at the HRR level.Patients were censored if they died, disenrolled from Medicare, or reached the end of the study period.Hazard ratios were cal-culated using marginal effect estimates from the Cox models.Operative recurrence numbers were estimated using the Cox models with covariates set to their mean values.Proportional hazards assumptions were tested using Schoenfeld residuals.For variables that violated this assumption, we included an interaction term with the logarithm of time in the second stage of our model. 25Variables that required an interaction term with time included surgical approach, age, sex, chronic pulmonary disease, hypothyroidism, kidney failure, fluid and electrolyte disorders, and obesity.
To assess the strength of the IV, we first calculated the Kleibergen-Paap Wald F statistic for prior-year use of roboticassisted ventral hernia repair and the current-year treatment (eg, undergoing robotic-assisted ventral hernia repair).Our F statistic of 63 966 demonstrated that our IV was highly associated with undergoing robotic-assisted ventral hernia, as an F statistic greater than 10 is generally considered to be a strong instrument. 26A valid instrument must not be associated with the outcome except through the treatment variable.While this condition cannot be empirically proven, it can be evaluated on both a theoretical basis and by examining the balance of patient characteristics when stratified by the instrument.For the former theoretical reasoning, local lagged treatment patterns are believed to satisfy this condition, as they reflect clinician treatment decisions from a prior time period among a different set of patients. 27Additionally, regional use of roboticassisted surgery in a previous year is highly likely to influence its use in the subsequent year.For the latter stratification, we analyzed baseline patient characteristics for as-treated cohorts and for patients stratified around the median of our IV (eTables 2 and 3 in Supplement 1).With a strong instrument, Surgical approach Robotic-assisted Open Laparoscopic Cumulative incidence of operative hernia recurrence was calculated using a Cox proportional hazards model that adjusted for patient age, sex, race and ethnicity, Elixhauser comorbidities, year of surgery, approach (robotic-assisted, laparoscopic, or open), mesh use, the use of myofascial flap, and procedure type (ventral/incisional or umbilical).Analysis included use of an instrumental variable to reduce measured and unmeasured confounding.The instrument used was robotic-assisted ventral hernia repair use rate within a hospital referral region in the 12 months prior to a patient's initial ventral hernia repair.95% CIs are not visible as the largest interval was −0.18% to 0.18% from point estimates.
patient-level covariates are ideally more similar across approaches when comparing the above-median and belowmedian levels of the instrument to the actual treatment level.
There was a significant reduction in covariate imbalance when evaluating by above-median and below-median groups of the IV vs the as-treated operative approach (eTables 2 and 3 in Supplement 1).For example, a standardized difference of 8.3% in the prevalence of chronic pulmonary disease between actual treatment groups (robotic-assisted: 2683 [21.1%]; laparoscopic: 8014 [24.6%]) was reduced to 0.1% using the instrument (robotic-assisted: 18 841 [23.4%]; laparoscopic: 18 999 [23.5%]).After implementation of the instrument, the number of covariates with a standardized difference greater than 10% was reduced compared to the as-treated covariates for robotic-assisted vs laparoscopic and robotic-assisted vs open approaches.Thus, we used an IV for our main analysis.
Several preplanned sensitivity analyses were performed in an identical manner restricting the patient population by hernia subtype (either ventral/incisional or umbilical), presence of obesity, number of comorbidities, ICD-9 and ICD-10 coding status, surgeon volume of robotic-assisted ventral hernia repairs, and the proportion of an individual surgeon's ventral hernia cases performed with robotic assistance.An additional sensitivity analysis was performed post hoc looking at myofascial flap use.Instrumental variable models were also run with clustering at either the hospital or surgeon level; however, this did not appreciably affect the CIs of our estimates.All analyses were performed using SAS version 9.4 (SAS Institute) and Stata version 15 (StataCorp LLC).Tests were 2-sided and significance was set at P < .05.Baseline unadjusted patient characteristics were compared across the 3 approaches using analysis of variance, Kruskal Wallis, or Pearson χ 2 tests as appropriate.The IV analysis used robust standard errors to account for HRR-level heteroscedasticity.Analysis was performed from January 2023 through March 2024.

Sensitivity Analyses
We performed multiple sensitivity analyses to test our results (Table 3).The previously mentioned findings were consistent across patients with obesity vs patients without obesity, patients with a low vs high comorbidity burden, when looking at ICD-9 vs ICD-10 coding, and for those without myofascial flap repair.For those hernia repairs that included myofascial flaps, open repair had lower 10-year operative recurrence rates while laparoscopic repair had higher recurrence rates than robotic-assisted repair.Robotic-assisted repair was associated with higher operative recurrence rates compared with the other 2 approaches across varying levels of surgeon robotic-assisted ventral hernia repair volume, although the magnitude of these differences was smaller for surgeons with the highest proportion of robotic-assisted ventral hernia repairs (ie, in the top 25%).

Discussion
In this analysis of Medicare beneficiaries from 2010 to 2020, the incidence of operative recurrence after ventral hernia repair was higher following robotic-assisted surgery compared with laparoscopic or open surgery.It is important to note that our findings should be interpreted within the context of evidence suggesting operative recurrence rates underestimate true clinical hernia recurrence by as much as 4 to 5 times. 28These results were consistent across different patient characteristics, hernia subtype (ventral/incisional or umbilical), and even among the highest-volume surgeons performing roboticassisted ventral hernia repairs in the sample.These data suggest robotic-assisted ventral hernia repair has inferior longterm outcomes compared with established open and laparoscopic approaches, calling into question the clinical rationale behind its widespread and rapid adoption over the last decade.
The results of this study are complementary to and consistent with prior randomized clinical trials and observational studies.For example, a similar study from 2022 demonstrated higher rates of operative recurrence after minimally invasive vs open ventral hernia repair without differentiating between laparoscopic and robotic-assisted approaches. 21Our work expands on this finding by explicitly comparing operative recurrence between all 3 approaches.Our results conflict with earlier work in the same Medicare population from 2007 to 2015, which found fewer recurrences after robotic- Surgical approach Robotic-assisted Open Laparoscopic Operative recurrence following umbilical hernia repair

B
Cumulative incidence of operative hernia recurrence following ventral or incisional (A) and umbilical (B) hernia repair was calculated using a Cox proportional hazards model that adjusted for patient age, sex, race and ethnicity, Elixhauser comorbidities, year of surgery, approach (robotic-assisted, laparoscopic, or open), mesh use, the use of myofascial flap, and hernia subtype (ventral/incisional or umbilical).Analysis included use of an instrumental variable to reduce measured and unmeasured confounding.The instrument used was robotic-assisted ventral hernia repair use rate within a hospital referral region in the 12 months prior to a patient's initial ventral hernia repair.95% CIs are not visible as the largest interval was −0.18% to 0.18% from point estimates for ventral and incisional subtype hernias and −0.14% to 0.14% from point estimates for umbilical-subtype hernias.

Flap use
Robotic-assisted  20 These differences are potentially due to our methods, as we used an IV analysis to address selection bias and other unmeasured sources of confounding inherent in observational data.Moreover, rates of robotic-assisted ventral hernia repair increased sharply starting in 2014, and our study captures the diffusion of this approach out of specialized centers and into the broader surgical community.Current trends suggest that the use of robotic assistance will continue to grow, even without evidence of clinical superiority to other approaches.A 2022 study revealed that the major drivers of robotic-assisted hernia repair were not patient or hernia characteristics but rather market competition and availability of the robotic console. 295][6][7] While evidence to support these proposed advantages remains mixed, [30][31][32] our study highlights the potential downside for patients when demand for a new technology outpaces the availability of highquality evidence.
One area where robotic assistance may offer a clinical advantage is in repairing large or complex hernias with techniques that could not otherwise be performed laparoscopically.For example, robotic-assisted component separation, a commonly used abdominal wall reconstructive technique (coded as myofascial flap in claims data), is associated with shorter length of stay and fewer complications compared to open repair. 33However, most surgeons across the country are not using robotic assistance for these complex repairs, which are typically performed by only the most experienced abdominal wall surgeons.In our study, higher surgeon roboticassisted ventral hernia repair volume did lessen the magnitude of differences in operative recurrence rates, but did not eliminate differences entirely.Thus, experience may be a crucial component to improving outcomes with robotic-assisted hernia repair, and the ability to right-size adoption necessitates improved training and credentialing that align with the modern realities of surgical practice.
Increases in the incidence and complexity of abdominal wall hernias have led to a push for hernia-specific fellowship training. 34While this may improve the pipeline of surgeons with hernia expertise, it does not address the broader community of general surgeons who perform most hernia repairs in the US.Various institution-level robotic training curricula have been implemented in surgical residencies across the country, but the American Board of Surgery does not require standardized robotic training or assessment for surgical trainees.This has produced widely varying training components, case exposure, and documentation of competency on residency graduation. 35Furthermore, access to robotic simulators and adequate case volumes during training is largely institution dependent, producing additional barriers to personalized trainee competency assessment. 36For surgeons who desire roboticassisted surgical training postresidency, credentialing requirements typically involve completion of an industrysponsored training course and performance of a certain number of proctored robotic-assisted cases. 37A 2021 study found these credentialing processes inadequate to determine proficiency due to varying requirements, lack of alignment of case numbers with procedure-specific learning curves, and an absence of outcome monitoring. 38Standardized training and credentialing for robotic-assisted surgery would provide a foundation on which to better understand a surgeon's proficiency and safety prior to using the technology in independent practice.

Limitations
Our work must be understood in the context of several limitations.First, the Medicare population may not be generaliz- able to all patients with hernias.However, ventral hernia repair is more common in older adults due to both aging and the higher likelihood of developing a hernia from a prior abdominal operation. 39Second, Medicare claims data are not able to capture nuanced clinical information, such as hernia size or mesh type.However, as most surgeons are not abdominal wall specialists, we would expect Medicare claims to reflect an average general surgical practice, where most robotic-assisted hernia repairs are performed on patients with smaller or less complex ventral hernias.Third, selection bias is a known limitation of observational studies; our study was designed specifically to address this issue by using an IV analysis.We confirmed mathematically that this instrument was strong and conformed to the standards of exogeneity.Finally, it is possible that our study period captures the global learning curve that accompanies the use of a new technique, similar to the early studies that showed laparoscopic ventral hernia repairs had higher recurrence rates than open repairs. 40While the dif-ferences we found may diminish once robotic assistance is used more universally, the incremental value added by the robot would remain in question.Even in the presence of similar outcomes, laparoscopy would still have the advantage of lower costs and resource use compared with robotic-assisted ventral hernia repair.

Conclusions
In this population-level observational study that used an IV analysis to control for unmeasured confounding, patients undergoing robotic-assisted ventral hernia repair had higher rates of operative recurrence than those undergoing laparoscopic or open repair up to 10 years after surgery.Narrower clinical applications that focus on the specific advantages and disadvantages of each approach may improve patient outcomes following ventral hernia repairs.

Figure 2 .
Figure 2. Cumulative Incidence of Operative Recurrence Following Ventral or Incisional Subtype Hernia Repairs Only and Umbilical Hernia Subtype Repairs Only, Stratified by Approach (Robotic-Assisted, Laparoscopic, and Open) From 2010-2020

Table 1 .
20,21t Characteristics by Operative Approach robotic-assisted, laparoscopic, or open), mesh use, the use of myofascial flap, and hernia subtype (ventral/incisional or umbilical).20,21Patientrace and ethnicity are included as identifiers in the Medicare claims database in this study and are represented in the following standard, fixed categories: Asian, Black, Hispanic, North American Native, White, other race, and unknown race.Only a single category could be selected, and no additional information is available in Medicare claims data regarding the composition of the other race category.Age was treated as a continuous variable.All other variables were treated as categorical.
a P values shown are for analysis of variance, Kruskal-Wallis, or Pearson χ 2 tests depending on variable type.b Comorbidities with a sample prevalence of at least 2% reported.c Mesh use includes any open procedure with a separate billing claim for mesh placement or any minimally invasive procedure (robotic, laparoscopic), as mesh placement is included in the billing claim for these case.Surgical Approach and Long-Term Recurrence After Ventral Hernia Repair Original Investigation Research jamasurgery.com(Reprinted) JAMA Surgery Published online June 12, 2024 E3 Downloaded from jamanetwork.comby guest on 06/29/2024 (

Table 3 .
Sensitivity Analyses With Hazard Ratios (HRs) for Operative Recurrence at 1, 3, 5, 7, and 10 Years After Robotic-Assisted, Laparoscopic, and Open Ventral Hernia Repair From 2010-2020 (continued)Abbreviations: ICD-9, International Classification of Diseases, Ninth Revision; ICD-10, International Classification of Diseases, Tenth Edition; NA, not applicable.a Myofascial flap analysis was performed using ventral or incisional subtype hernias only, as the technique is most commonly used for large or complex hernias and the number of umbilical hernias with myofascial flap composed less than 1% of all hernias coded with myofascial flap use.b Only 5 years of data were available for ICD-9 and ICD-10 sensitivity analyses.