The development and validation of risk-stratification models for short-term outcomes following contaminated complex abdominal wall reconstruction

Abstract

Background
Short-term outcomes for patients undergoing contaminated complex abdominal wall reconstruction (CCAWR), including risk stratification, have not been studied in sufficiently high numbers. This study aims to develop and validate risk-stratification models for Clavien–Dindo (CD) grade ≥ 3 complications in patients undergoing CCAWR.MethodsA consecutive cohort of patients who underwent CCAWR in two European national intestinal failure centers, from January 2004 to December 2015, was identified. Data were collected retrospectively for short-term outcomes and used to develop risk models using logistic regression. A further cohort, from January 2016 to December 2017, was used to validate the models.ResultsThe development cohort consisted of 272 procedures performed in 254 patients. The validation cohort consisted of 114 patients. The cohorts were comparable in baseline demographics (mean age 58.0 vs 58.1; sex 58.8% male vs 54.4%, respectively). A multi-variate model including the presence of intestinal failure (p < 0.01) and operative time (p < 0.01) demonstrated good discrimination and calibration on validation. Models for wound and intra-abdominal complications were also developed, including pre-operative immunosuppression (p = 0.05), intestinal failure (p = 0.02), increasing operative time (p = 0.04), increasing number of anastomoses (p = 0.01) and the number of previous abdominal operations (p = 0.02). While these models showed reasonable ability to discriminate patients on internal assessment, they were not found to be accurate on external validation.Conclusion
Acceptable short-term outcomes after CCAWR are demonstrated. A robust model for the prediction of CD ≥ grade 3 complications has been developed and validated. This model is available online at www.smbari.co.uk/smjconv2.

Full text

Vol.:(0123456789)
1 3
Hernia
https://doi.org/10.1007/s10029-019-02120-6
ORIGINAL ARTICLE
The development andvalidation ofrisk‑stratication models
forshort‑term outcomes followingcontaminated complex abdominal
wall reconstruction
J.D.Hodgkinson1,2· F.E.E.deVries3· J.J.M.Claessen3· C.A.Leo1,2· Y.Maeda1,2· O.vanRuler4· O.Lapid5·
M.C.Obdeijn5· P.J.Tanis3· W.A.Bemelman3· J.Constantinides1· G.B.Hanna2· J.Warusavitarne1,2·
M.A.Boermeester3· C.Vaizey1,2
Received: 1 October 2019 / Accepted: 30 December 2019
© Springer-Verlag France SAS, part of Springer Nature 2020
Abstract
Background Short-term outcomes for patients undergoing contaminated complex abdominal wall reconstruction (CCAWR),
including risk stratification, have not been studied in sufficiently high numbers. This study aims to develop and validate risk-
stratification models for Clavien–Dindo (CD) grade ≥ 3 complications in patients undergoing CCAWR.
Methods A consecutive cohort of patients who underwent CCAWR in two European national intestinal failure centers, from
January 2004 to December 2015, was identified. Data were collected retrospectively for short-term outcomes and used to
develop risk models using logistic regression. A further cohort, from January 2016 to December 2017, was used to validate
the models.
Results The development cohort consisted of 272 procedures performed in 254 patients. The validation cohort consisted
of 114 patients. The cohorts were comparable in baseline demographics (mean age 58.0 vs 58.1; sex 58.8% male vs 54.4%,
respectively). A multi-variate model including the presence of intestinal failure (p < 0.01) and operative time (p < 0.01)
demonstrated good discrimination and calibration on validation. Models for wound and intra-abdominal complications were
also developed, including pre-operative immunosuppression (p = 0.05), intestinal failure (p = 0.02), increasing operative
time (p = 0.04), increasing number of anastomoses (p = 0.01) and the number of previous abdominal operations (p = 0.02).
While these models showed reasonable ability to discriminate patients on internal assessment, they were not found to be
accurate on external validation.
Conclusion Acceptable short-term outcomes after CCAWR are demonstrated. A robust model for the prediction of
CD ≥ grade 3 complications has been developed and validated. This model is available online at www.smbar i.co.uk/smjco nv2.
Keywords Contaminated· Abdominal wall reconstruction· Risk stratification
Introduction
Complex abdominal wall reconstruction (CAWR) has
emerged rapidly over the last 20years with improved sur-
vival following abdominal catastrophe, advances in surgical
techniques and bioprosthetics. The Ventral Hernia Working
Group (VHWG) published guidelines in 2010, aiming to
risk stratify these patients by comorbidity and underlying
wound and operative contamination risk [1]. They grade
cases from 1 (clean defects in healthy patients) to 4 (con-
taminated defects in patients with other comorbidity). It has
been suggested by Kanters etal. [2], that combining grades
3 and 4, and stratifying contamination according to the
Centre for Disease Control (CDC) classification of wound
* J. D. Hodgkinson
jonathanhodgkinson@nhs.net
1 Department ofColorectal Surgery, St Mark’s Hospital,
Academic Institute, Watford Road, HarrowHA13UJ, UK
2 Department ofSurgery andCancer, Imperial College
London, London, UK
3 Department ofSurgery, Amsterdam University Medical
Centers, University ofAmsterdam, Amsterdam,
TheNetherlands
4 Department ofSurgery, IJsselland Ziekenhuis,
Capellea/DIJssel, TheNetherlands
5 Department ofPlastic andReconstructive Surgery,
Amsterdam University Medical Centers Amsterdam,
University ofAmsterdam, Amsterdam, TheNetherlands

Hernia
1 3
contamination scale improves the accuracy of prediction of
wound morbidity.
Short-term outcomes in patients undergoing reconstruc-
tion of contaminated (modified-VHWG 3) defects have
not been sufficiently studied in high numbers. A recently
published systematic review demonstrated high overall
wound complication rates [pooled-proportion 46% (range
18–85%)], which may be expected in a group of patients
with pre-operative active wound infection/contamination [3].
The pooled post-operative mortality rate of 2.3% was accept-
ably low. There was no standardization of outcome measures
reported across the studies and minimal use of the validated
Clavien–Dindo (CD) scoring system [4].
A number of studies have attempted to determine risk
factors for wound complications in this group of patients.
Factors such as active smoking, diabetes mellitus, chronic
obstructive pulmonary disease (COPD), number of previous
hernia repairs, number of previous abdominal operations,
and increasing operative time were found to be significant
predictors for wound complications [5–8]. These studies,
however, all used small cohorts (range 37–128) and are,
therefore, at risk of type one error. No analysis has been
performed in modified-VHWG grade 3 (contaminated) to
look for risk factors associated with other post-operative
comorbidity, such as CD grade or return to theatre.
One previous study has developed a risk-stratification
score for prediction of surgical site occurrence (SSO) and
surgical site infection (SSI) following ventral hernia repair
[9]. They successfully validate the score on an external
prospective cohort [10]. The most significant predictor of
SSO and SSI in this scoring system is wound contamination
and, therefore, of limited use to discriminate risk in patients
with modified-VHWG 3 defects. It can be hypothesized
that evaluating this cohort in isolation would lead to a more
accurate model of risk stratification, therefore, this study
aims to develop and validate risk-stratification models for
CD grade ≥ 3 post-operative complications for patients with
modified-VHWG grade 3 abdominal wall defects.
Methods
A collaboration was set up between St Mark’s Hospital,
London, UK (SMH) and the Amsterdam University Medi-
cal Centres, location AMC, Amsterdam, the Netherlands
(AUMC), two national referral centres for complex gastro-
intestinal surgery and intestinal failure. Ethical approval was
granted by the local institutional review board and national
ethics committee in both centres (UK: REC ref 16/EE/0348;
IRAS 210325). The initial cohort used to develop the model
was established from consecutive patients undergoing elec-
tive repair of a midline modified-VHWG grade 3 abdomi-
nal wall defect [2], between January 2004 and December
2015, through local operative databases. The presence of
contamination, to define Grade 3 cases, was taken as the
presence of a stoma, concomintant gastrointestinal tract vio-
lation, the presence of an infected mesh, septic dehiscence
or enterocutaneous fistula. To validate the models, an inde-
pendent cohort of consecutive patients was identified from
January 2016 to December 2017. At both centres, data were
collected via as retrospective review of prospectively col-
lected clinical databases.
Surgical technique
Both centres have a long-standing history of treating patients
with contaminated abdominal wall defects and intestinal
failure. All procedures were performed in the elective set-
ting and, therefore, patients received pre-operative medical
optimisation, including weight loss and diabetic control.
Operative strategies were comparable between the centres
and consistent with published guidelines and consensus at
the time of operation. Briefly, according to consensus on
the management of patients with intestinal failure [11],
enterocutaneous fistulas were resected and a primary hand
sewn anastomosis was usually performed. Proximal divert-
ing stomas were used when necessary. The primary aim of
repair of the abdominal wall defect was tension-free fascial
closure with reinforcing retro-rectus or intra-abdominal
mesh whenever possible. Open component separation tech-
niques (CST) were used if appropriate and possible. Plastic
and reconstructive surgeons were consulted and performed
reconstructions where deemed appropriate.
Fascial closure was typically performed using a looped
monofilament polydioxanone (PDS) suture. If fascial clo-
sure was not achievable, the defect was bridged using an
intra-peritoneal onlay mesh or retro-muscular (sublay) mesh.
Passive intra-abdominal drains were used where deemed
appropriate but not routinely, and active/closed suction
subcutaneous drains were used routinely. AUMC preferred
interrupted polyester Mersilene sutures (Ethicon, Johnson
and Johnson, Belgium), whereas SMH tended to use surgical
clips or interrupted prolene sutures for skin closure.
Patient demographics andoutcomes
Data were collected for baseline patient demographics
including sex, age and body mass index (BMI) at time of
surgery, comorbidity, number of previous abdominal opera-
tions and previous hernia repairs. Previous open abdomen
was defined as long-term management with laparostomy.
Operative details including operative time, technique of
abdominal wall reconstruction and concurrent procedures
were recorded. The validated CD grade of surgical com-
plications was used as the primary short-term outcome
measure [4]. Wound morbidity was divided into any wound

Hernia
1 3
complication, and the wound intervention required to treat
the complication (conservative/bedside measures vs radio-
logical or surgical intervention) was recorded. Similarly,
intra-abdominal septic complications (IASC) (anastomotic
leak, intra-abdominal collection and recurrent fistula) were
separately defined by their occurrence and required inter-
vention to treat the complication (conservative/bedside
measures vs radiological or surgical intervention). This is
consistent with recent recommendations in outcome report-
ing in this cohort [12].
Statistical analysis
Baseline characteristics are presented as proportions and
compared between cohorts using the Mann–Whitney test.
Uni-variate binary logistic regression analysis was used to
identify factors associated with CD ≥ 3, wound interven-
tion and IASC intervention within the development cohort.
Multi-variate binary logistic regression, using a backward
selection process, was then undertaken on factors of signifi-
cance to determine a risk-stratification model, with inclu-
sion significance level set at a p value of ≤ 0.05. The logit-p
equation was used to determine predicted probabilities for
each outcome using factors retained in multi-variate analysis
[log(p/1 − p) = y = constant + risk factor A (regression coef-
ficient) + risk factor B (regression coefficient) + ···], with the
probability of complication being defined as P = (ey/1 + ey)).
Receiver operating characteristics (ROC) curves were ana-
lysed to assess the predictive ability of the models.
Using the validation cohort, model discrimination was
tested by examining the difference in predictive probability
of a complication when it occurred compared to when it did
not. The predicted probabilities were non-parametric and
compared using the Mann–Whitney test. Discrimination
was then further evaluated using the c-statistic with a 95%
confidence interval. Model calibration was evaluated using
the Hosmer–Lemeshow test to compare the predicted and
observed number of adverse events. Differences between
these groups were assessed using a Chi-squared test with
one degree of freedom. Both discrimination and calibration
were evaluated together by dividing patients into three cat-
egories. These categories were based on the predicted risk
of an outcome parameter, and by comparing the observed
percentage of an outcome with the predicted risk.
Results
Development cohort: baseline patient
demographics andoutcomes
The development cohort consisted of 269 consecutive
patients who underwent 287 procedures. Fifteen patients
who had incomplete or missing records were excluded from
the analysis. This resulted in 272 procedures in 254 patients
included in the final analysis. The mean age was 58.0 (SD
13.6years) and 58.8% were male (Table1).
The median post-operative length of stay was 18days
(IQR 10–31). Eighty-four percent of patients had a post-
operative complication on the CD scale, of which 43.8%
were of grade 3 or higher. A wound complication was
recorded in 58.5% and an intra-abdominal septic complica-
tion (see “Methods”) in 18.8% (Table2).
Validation cohort: baseline demographics
andoutcomes
The validation cohort consisted of 112 consecutive patients
who underwent 114 procedures. The mean age was 58.1
(SD 12.6years) and 54.4% were male (Table1). No signifi-
cant differences were seen between the pre-operative demo-
graphics of the validation and development cohorts, with
the exception of the number of patients with a history of
treatment with an open abdomen (p = 0.01) and the number
where fascial closure was achieved (p = 0.01).
The average length of stay was 19days (IQR 10–37),
which was comparable to the development cohort. Any com-
plication on the CD scale was seen in 80.7% of patients,
with 40.2% having a complication of grade 3 or higher. A
wound complication was recorded in 50.9% and an IASC in
23.7% (Table2).
Model 1: Clavien–Dindo complication grade 3
ormore
Uni-variate analysis identified a number of factors associ-
ated with CD ≥ 3. Patient factors included cardiac comor-
bidity, previous treatment with open abdomen, presence of
intestinal failure and presence of an enterocutaneous fistula.
Operative factors were operative time, number of bowel
anastomoses, the need for component separation techniques
and the inability to obtain fascial closure (Table3).
The factors found to have a significant effect on the like-
lihood of a CD ≥ 3 complication were entered into a back-
ward stepwise multi-variate logistic regression model. In
the multi-variate model, the presence of intestinal failure
(p < 0.01) and total operative time (p < 0.01) remained sig-
nificant and were included in the final risk model (Table4).
Evaluation of the model using receiver operating characteris-
tics (ROC) revealed a c-statistic of 0.72 (95% CI 0.66–0.79),
suggesting a reasonable ability of the model to discriminate
between high- and low-risk patients.
Assessment of the models discriminative ability was per-
formed on the validation cohort. A comparison of the predic-
tive percentage of the model revealed a p value of < 0.0001,
suggesting that the model is able to discriminate between

Hernia
1 3
patients. ROC assessment of the model on the validation
cohort demonstrated a c-statistic of 0.73 (95% CI 0.63–0.82)
(Fig.1). Analysis of the model calibration revealed evidence
of good calibration as the predicted and observed complica-
tion rates were comparable Examination of this using the Hos-
mer–Lemeshow test showed a p value of 0.12, consistent with
good calibration of the model for this outcome (Table5).
Model 2: wound intervention
Uni-variate analysis demonstrated patient factors including
the presence of intestinal failure, previous treatment with
an open abdomen and the number of previous abdominal
operations ≥ 3 as associated with the requirement for wound
intervention. Operative factors were also significantly asso-
ciated with wound intervention, such as increasing operative
time, the number of anastomoses and the use of component
separation techniques (Table3).
Multi-variate analysis retained pre-operative immunosup-
pression (p = 0.05), intestinal failure (p = 0.02), increasing
operative time (p = 0.04) and increasing number of anas-
tomoses (p = 0.01) in the final model. ROC demonstrated
a reasonable ability of the model to discriminate between
high- and low-risk patients (c = 0.76; 95% CI 0.68–0.83)
(Table4).
On assessment of the model using the validation cohort,
poor discrimination of the model was demonstrated (Fig.1).
Table 1 Baseline demographics of the development and validation cohorts
Differences between the cohorts have been evaluated using the Mann–Whitney test (α = 0.05)
BMI body mass index, ASA American Society of Anaesthesiology, COPD chronic obstructive pulmonary disease, IBD inflammatory bowel dis-
ease SD standard deviation, IQR inter-quartile range
Development cohort Validation cohort p value
272 procedures in 254 patients 114 procedures in 112
patients
Patient characteristics
Age, mean (SD) 58.0 (13.6) 58.1 (12.6) 0.94
Sex male, n (%) 160 (58.8) 62 (54.4) 0.42
BMI, median (IQR) 26.0 (22.6–29.6) 26.0 (22.9–30.5) 0.59
ASA classification, mean (SD) 2.43 (0.5) 2.45 (0.6) 0.70
1 0 (0) 3 (2.6)
2 162 (59.6) 61 (53.5)
3 103 (37.9) 46 (40.4)
4 7 (2.6) 4 (3.5)
Active smoker, n (%) 62 (22.8) 16 (14.0) 0.06
Diabetes, n (%) 50 (18.4) 16 (14.0) 0.43
Immunosuppression, n (%) 21 (7.7) 9 (7.9) 0.96
Cardiac comorbidity, n (%) 63 (23.2) 28 (24.6) 0.77
Pulmonary comorbidity, n (%) 55 (20.2) 34 (29.8) 0.04
COPD, n (%) 29 (10.7) 11 (9.6) 0.75
Hypertension, n (%) 83 (30.5) 34 (29.8) 0.89
Abdominal malignancy, n (%) 50 (18.4) 31 (27.2) 0.06
IBD, n (%) 39 (14.3) 24 (21.1) 0.06
Intestinal failure, n (%) 129 (47.1) 48 (42.1) 0.47
History of open abdomen, n (%) 128 (47.1) 38 (33.3) 0.01
Presence of intestinal fistula, n (%) 159 (58.1) 60 (52.6) 0.42
Operative characteristics
Number of previous abdominal surgeries, median (IQR) 4 (2–5) 3 (2–5) 0.18
Undergone previous hernia repair/s, n (%) 119 (43.8) 49 (43.0) 0.90
Anastomosis constructed, n (%) 200 (73.5) 84 (73.7)
Mesh removal, n (%) 59 (21.7) 23 (20.2) 0.74
Open component separation technique performed, n (%) 183 (67.3) 85 (74.6) 0.20
Mesh used, n (%) 182 (66.9) 77 (67.5) 0.90
Fascial closure achieved, n (%) 197 (72.5) 99 (86.8) 0.01

Hernia
1 3
The predicted percentages of patients with and without a
wound intervention were not significantly different on
Mann–Whitney testing (p = 0.61), and evaluation of this
data with the c-statistic revealed a value of 0.54 (95% CI
0.36–0.73). Calibration testing also did not demonstrate ade-
quate calibration of the model with poor correlation between
predicted and observed interventions and a significant differ-
ence between the observed numbers and the predicted values
(p < 0.001) (Table5).
Model 3: intra‑abdominal septic complications
Analysis with uni-variate regression revealed only five fac-
tors to be associated with an IASC: pre-operative immu-
nosuppression; pre-operative intestinal failure; previous
treatment with an open abdomen; presence of an enterocuta-
neous fistula and a number of previous abdominal operations
of three or more (Table3). These factors were, therefore,
entered into multi-variate analysis. Pre-operative immu-
nosuppression (p = 0.04), pre-operative intestinal failure
(p = 0.02) and the number of previous abdominal opera-
tions (p = 0.02) remained in the final model. ROC revealed
a reasonable ability of the model to predict this outcome
(c = 0.72; 95% CI 0.64–0.80) (Table4).
Assessment of the model using the validation cohort
revealed no significant difference in the predicted percent-
age of intervention versus no intervention (p = 0.62) (Fig.1).
Evaluation of this result using the c-statistic revealed a value
of 0.47 (95% CI 0.32–0.61) suggesting poor ability of the
model to discriminate between low- and high-risk patients.
On assessment of the calibration of the model, there was
no observed consistent pattern between the observed and
predicted rate of intervention in each group. This revealed
that the model significantly under-estimated the rate of inter-
vention in the low-risk group. Hosmer–Lemeshow testing
showed a significant difference between the observed and
predicted number of interventions (p < 0.001) suggest-
ing poor calibration of the model to predict this outcome
(Table5). The significant difference observed suggests poor
calibration.
Discussion
This study aimed to evaluate short-term outcomes, and to
risk stratify patients undergoing contaminated (VHWG
grade 3) complex abdominal wall reconstruction. Develop-
ment and validation of a risk-stratification model for predict-
ing CD complications of grade 3 or more in CCAWR has
been demonstrated. This is an inherently heterogeneous and
complex group of patients as illustrated by the extensive
number of comorbidities, high mean ASA grade, high num-
ber of previous abdominal operations and inherent surgical
site contamination. Nevertheless, good short-term outcomes
with acceptable rates of wound complication and post-oper-
ative enterocutaneous fistula recurrence were found. While
the total number of patients who experienced any complica-
tion (84.9%) and the median length of stay are high, most
Table 2 Short-term post-
operative outcomes from the
development and validation
cohorts
IQR inter-quartile range
Outcome Development
(N = 272)
Validation (N = 114)
Length of stay (median days, IQR) 18 (10–31) 19 (10–37)
Clavien–Dindo complication, n (%)
0 41 (15.1) 22 (19.3)
1 47 (17.3) 14 (12.3)
2 65 (23.9) 33 (28.9)
3a 38 (14.0) 15 (13.2)
3b 19 (7.0) 8 (7.0)
4a 44 (16.2) 9 (7.9)
4b 11 (4.0) 11 (9.6)
5 6 (2.6) 3 (2.6)
Post-operative wound complication, n (%) 159 (58.5) 58 (50.9)
Wound intervention required (interventional radiology or
surgery), n (%)
71 (26.1) 14 (12.3)
Intra-abdominal septic complication, n (%) 51 (18.8) 27 (23.7)
Intra-abdominal septic complication intervention, n (%) 42 (15.4) 22 (19.3)
Return to theatre, n (%) 33 (12.1) 12 (10.5)
Re-admission within 30days of discharge, n (%) 35 (12.9) 6 (5.3)
Recurrent enterocutaneous fistula, n (%) 20 (7.4) 4 (3.5)

Hernia
1 3
complications were of low grade (CD1/2) and, therefore, this
is something that can be expected from such a complex type
of surgery [3, 13, 14].
The model for CD outcome appears to have good pre-
dictive ability, both in discrimination and calibration
tests. This is the first risk-stratification tool designed and
validated for use in contaminated ventral hernia surgery,
and the first that predicts the risk of a validated outcome
measure (Clavien–Dindo grade). Models predicting other
short-term outcomes (wound and IASC interventions)
Table 3 Uni-variate analysis demonstrating significant factors associated with three key outcomes
Bold highlighted factors demonstrate a p value of ≤ 0.05 and were used in multi-variate analysis
COPD chronic obstructive pulmonary disease, IBD inflammatory bowel disease, ECF enterocutaneous fistula, S/C subcutaneous, IA intra-
abdominal
Risk factor Clavien–Dindo Grade ≥ 3
(OR (95% CI))
Wound intervention (OR
(95% CI))
Intra-abdominal septic complication
(IASC) intervention (OR (95% CI))
Age 1.01 (0.99–1.03) 0.99 (0.97–1.01) 1.00 (0.97–1.02)
Sex (female) 1.10 (0.67–1.79) 1.24 (0.72–2.14) 0.97 (0.50–1.89)
Obese 0.92 (0.51–1.65) 0.93 (0.48–1.78) 0.77 (0.33–1.77)
Smoker 0.61 (0.33–1.11) 0.61 (0.30–1.23) 0.52 (0.21–1.30)
Diabetes 1.02 (0.55–1.91) 0.76 (0.37–1.58) 1.48 (0.68–3.26)
Immunosuppression 1.31 (0.54–3.19) 2.29 (0.92–5.68) 3.09 (1.16–8.18)
Cardiac comorbidity 2.29 (1.29–4.05) 0.76 (0.39–1.49) 1.22 (0.57–2.58)
Pulmonary comorbidity 1.61 (0.89–2.92) 0.85 (0.43–1.69) 1.09 (0.49–2.44)
COPD 1.59 (0.73–3.43) 0.89 (0.36–2.18) 1.16 (0.42–3.23)
IBD 0.44 (0.20–0.93) 0.83 (0.37–1.84) 0.78 (0.29–2.12)
Intestinal Failure 2.59 (1.58–4.25) 2.05 (1.18–3.57) 2.90 (1.43–5.87)
Pre-op resistant bacteria 1.43 (0.67–3.06) 0.83 (0.34–2.02) 1.09 (0.39–3.02)
N. previous abdominal surgeries ≥ 3 2.31 (1.27–4.19) 3.00 (1.39–6.42) 5.27 (1.57–17.67)
N. previous hernia surgeries
1 1.41 (0.81–2.47) 0.86 (0.47–1.61) 0.97 (0.47–2.01)
≥ 2 0.61 (0.30–1.26) 0.38 (0.15–0.95) 0.22 (0.05–0.97)
Recurrent hernia 0.95 (0.58–1.55) 0.57 (0.32–1.00) 0.64 (0.32–1.28)
Previous component separation 1.48 (0.59–3.69) 0.95 (0.33–2.72) 0.60 (0.13–2.69)
Previous open abdomen 1.71 (1.05–2.79) 1.94 (1.12–3.36) 2.30 (1.16–4.55)
Wound classification
Contaminated 1.85 (1.04–3.30) 2.18 (1.13–4.24) 3.57 (1.49–8.57)
Dirty 1.41 (0.72–2.76) 0.92 (0.40–2.10) 0.69 (0.19–2.47)
ECF 1.76 (1.06–2.90) 1.69 (0.96–2.99) 2.60 (1.22–5.53)
Stoma 1.14 (0.70–1.90) 0.84 (0.48–1.46) 1.33 (0.66–2.70)
Time from last operation 1.78(0.75–4.23) 2.16 (0.72–6.49) 2.44 (0.56–10.71)
Operation time
3.5–5.5h 0.93 (0.43–2.03) 0.91 (0.37–2.26) 0.66 (0.17–2.59)
5.5–7.5h 2.02 (0.95–4.26) 2.13 (0.93–4.92) 3.40 (1.16–9.97)
≥ 7.5h 9.15 (3.62–23.1) 4.48 (1.73–10.1) 4.85 (1.59–14.81)
Mesh removal 0.96 (0.53–1.72) 0.58 (0.28–1.20) 0.69 (0.29–1.63)
Number of anastomosis
1 1.58 (0.78–3.21) 3.22 (1.07–9.65) 1.77 (0.57–5.50)
≥ 2 3.46 (1.55–7.69) 10.0 (3.21–31.2) 4.40 (1.38–14.04)
Stoma takedown/relocation 1.42 (0.87–2.31) 1.40 (0.82–2.42) 1.46 (0.75–2.82)
Anterior component separation 1.85 (1.08–3.18) 2.79 (1.41–5.54) 2.15 (0.95–4.87)
Drain use 1.71 (0.86–3.39) 1.29 (0.60–2.75) 4.60 (1.07–19.77)
S/C drain 1.90 (1.00–3.62) 1.48 (0.72–3.07) 2.00 (0.75–5.37)
IA drain 1.20 (0.74–1.94) 1.26 (0.74–2.17) 1.56 (0.80–3.03)
Fascial closure 1.83 (1.06–3.17) 1.58 (0.87–2.87) 1.72 (0.85–3.48)

Hernia
1 3
Table 4 Risk-stratification models for Clavien–Dindo complications ≥ 3, wound intervention, IASC intervention and further hernia surgery developed using multi-variate backward selection
logistic regression
c-statistics derived using receiver-operating characteristic demonstrate values all greater than 0.70 suggesting a moderate discriminatory ability of all models
IASC intra-abdominal septic complication
Outcome Variable Category Odds ratio (95% CI) Regression coefficient p-value c-statistic
Clavien Dindo ≥3
Significance 0.05
Removal 0.05
Inclusion 0.05
Constant – – −1.80 – 0.72 (0.66–0.79)
Intestinal Failure No 1 <0.01
Yes 3.08 (1.56–6.06) 1.12
Operation time ≤3.5h 1 <0.01
3.5–5.5h 1.34 (0.53–3.35) 0.29
5.5–7.5h 3.53 (1.46–8.52) 1.26
>7.5h 11.39 (3.92–33.16) 2.43
Wound Intervention
Significance 0.05
Removal 0.05
Inclusion 0.05
Constant – – −3.02 – 0.76 (0.68–0.83)
Immuno-suppression No 1 0.05
Yes 3.02 (0.98–9.29) 1.11
Intestinal Failure No 1 0.02
Yes 2.20 (1.11–4.37) 0.79
Operation time ≤ 3.5h 1 0.04
3.5–5.5h 0.88 (0.29–2.72) −0.12
5.5–7.5h 2.20 (0.78–6.24) 0.79
>7.5h 3.17 (1.06–9.43) 1.15
Number of anastomosis 0 1 0.01
1 2.41 (0.63–9.21) 0.88
2+ 6.04 (1.46–25.0) 1.80
IASC Intervention
Significance 0.05
Removal 0.05
Inclusion 0.10
Constant – – −3.60 – 0.72 (0.64–0.80)
Immuno-suppression No 1 0.04
Yes 2.86 (1.03–7.90) 1.05
N. previous abdominal
surgeries
≤ 2 1 0.02
3+ 4.36 (1.28–14.83) 1.47
Intestinal Failure No 1 0.01
Yes 2.61 (1.26–5.41) 0.96

Hernia
1 3
have not demonstrated good predictive ability on vali-
dation. Previous studies in smaller cohorts have demon-
strated some of the factors identified in these models to
also be significant. The choice of wound and abdominal
intervention as outcome measures has been selected based
on giving the maximal clinical ability to counsel patients
pre-operatively. For example, the ability to predict and
inform a patient they have high risk for a wound compli-
cation in a cohort of patients was a 50% chance of having
a wound complication is of limited use [3]. The aim of
these models is to aid in discrimination by defining a risk
for a given patient of requiring surgical or radiological
intervention for wound complications compared to simple
measures, such as oral antibiotics. It is unclear why these
models proved less effective on validation in the present
study. However, it is possibly due to the smaller size of
the validation cohort or related to changes in practice pat-
terns, such as the introduction of closed incision nega-
tive pressure wound therapy, which potentially reduced
the rate of wound infection [15]. This was necessary to
include patients and to mount sufficient numbers due to the
available length of follow-up. Attempts were undertaken
to validate the models, however, additional centres with
adequate volume and a similar case mix could not be iden-
tified. As the prevalence of these cases at our two institu-
tions has increased in recent years, a further 2–3years of
data collection should provide the evidence to validate or
redevelop the current models.
The Ventral Hernia Risk Score (VHRS) is currently the
best attempt to risk stratify patients undergoing ventral
hernia surgery [9]. This score also includes contaminated
wound class as an independent predictor. However, if a
patient has a contaminated wound, this immediately places
them into the highest risk bracket, as defined by the score,
despite the presence or absence of any other risk factor. This,
therefore, limits its ability to discriminate in our cohort of
patients. The currently developed models aim to fill this gap.
The principle limitation of this study is the retrospective
nature of the majority of the data, particularly in the devel-
opment cohort. Every effort has been made to ensure the
robustness of the data collection including cross-checking
case notes and clinic letters, review by independent data col-
lectors and discussion with patients when they attended for
prospective follow-up. The age of some of this data also
limits effectiveness.
The inherently heterogeneous nature of clinical factors
within this cohort of patients, such as the range of reasons
for underlying contamination, range of patient comorbidity,
variety of repair techniques, mesh used and changing prac-
tice patterns over time also limits the analysis. Designing
a prospective study, controlling for all of these factors and
generating sufficient patient numbers for this analysis would
be impractical. We have attempted to control for as much of
Fig. 1 Evaluation of the discriminative ability of the models using
predictive probability of a given outcome. A significant difference
was seen between predicted groups when using the Clavien–Dindo
model suggesting the model has a moderate ability to discriminate
between patients. The other two models did no demonstrate a signifi-
cant difference

Hernia
1 3
the inherent difference as possible by standardising inclu-
sion criteria and repair decisions based on the best available
current guidelines.
Another limitation lies within the use of regression analysis
and is affected by the paucity of previous analysis of risk fac-
tors in this cohort. The ideal methodology for regression anal-
ysis would use previously identified risk factors and perform
multi-variate analysis to identify the overall effect of each fac-
tor, as well as analysis of factors that should be included in the
risk model. To avoid over-fitting the model, a factor can only
be included in the multi-variate analysis for every ten events
for that outcome occurring in the dataset. Due to the lack of
previous analysis of risk factors in this cohort, we have used
uni-variate analysis to identify significant variables to include
Table 5 A table demonstrating
calibration of the models using
Hosmer–Lemeshow and Chi-
squared tests
Model/risk group Number of
patients
Observed number of
complications (%)
Predicted number of
complications (%)
Chi-squared
statistic (p
value)
Clavien–Dindo ≥ 3
≤ 25% 42 10 (23.8) 8.5 (20.4) 2.5 (0.12)
26–50% 35 12 (34.3) 13.5 (38.5)
≥ 50% 36 23 (63.9) 26.6 (74.0)
Wound intervention
≤ 15% 43 5 (11.6) 3.4 (8.0) 19.7 (< 0.001)
16–30% 39 5 (12.8) 8.4 (21.4)
≥ 30% 31 4 (12.9) 15.5 (50.1)
IASC intervention
≤ 10% 34 9 (26.5) 1.6 (4.6) 37.2 (< 0.001)
11–20% 41 4 (9.8) 4.2 (10.8)
≥ 20% 39 9 (23.1) 9.6 (24.5)
Fig. 2 An example of the Convention webtool in use

Hernia
1 3
in the multi-variate analysis, which puts the model at risk of
over-fitting. However, the use of subsequent backward selec-
tion and good area under the curve minimise this and improve
the chances of model validation. This is demonstrated in the
successful validation of the model for CD ≥ 3.
Acceptable short-term outcomes have been demonstrated
in this large cohort of contaminated complex abdominal wall
reconstructions. The reported model provides the ability to
accurately predict the likelihood of a CD ≥ 3 outcome in these
patients. While a number of these risk factors have been iden-
tified previously and would be discussed with patients, this
tool will likely improve the ability of the surgeon to counsel
patients pre-operatively, by increasing the accuracy of the
strength of the effect of each risk factor and more precisely
determining a specific patients risk in an inherently heterog-
enous cohort. The tool could also improve a patients’ outcome
by better facilitation of resource planning for the surgery, and
checking of the availability of high-dependence post-operative
care for high-risk patients. This model is available online at
www.smbar i.co.uk/smjco nv2 for use by clinicians (Fig.2).
Given the rarity of the condition, the only way to prospec-
tively evaluate short- and long-term outcomes is through
collaboration and national/international database develop-
ment. The identification of significant risk factors in this
study could provide the basis of what factors should be
stored in such a database.
Acknowledgements The authors would like to acknowledge Mr Paul
Bassett, of StatsConsultancy, for his help and support with statistical
analysis and model development. The authors would like to thanks Mr
Victor Lesk for his work in developing the Convention webtool.
Funding A research award from the British Hernia Society kindly
funded the development and creation of the model webtool calculator.
Compliance with ethical standards
Conflict of interest JDH, FEEdV, JJMC, CAL, YM, OvR, OL, MCO,
PJT, WAB, JC, GBH and JW declare no conflict of interest directly re-
lated to the current work; MAB declares no conflict of interest directly
related to the current work and reports institutional research grants
from Baxter, Mylan, Ipsen, Acelity/KCI, Bard, LifeCell and Johnson
& Johnson/Ethicon and New Compliance; and is a speaker or advisory
board member for Acelity/KCI, Bard, LifeCell/Allergan, Gore, Bard,
Smith&Nephew and Johnson & Johnson / Ethicon; CJV declares no
conflict of interest directly related to the current work and declares
consultancy advisor to Acelity and paid lecture for Allergan.
Ethical approval This study was performed in accordance with the
ethical standards of the institutional and national research committee
(East of England REC ref 16/EE/0348; IRAS 210325) and with the
1964 Helsinki Declaration and its later amendments or comparable
ethical standards. Ethical approval for the study was given by local and
national bodies in both centres as stated in the manuscript.
Human and animal rights This retrospective study involving human
participants was in accordance with the ethical standards of the insti-
tutional and national research committee.
Informed consent Data were obtained from records obtained for clini-
cal purposes.
References
1. VHW Group, Breuing K, Butler CE, Ferzoco S, Franz M, Hult-
man CS etal (2010) Incisional ventral hernias: review of the lit-
erature and recommendations regarding the grading and technique
of repair. Surgery. 148(3):544–558
2. Kanters AE, Krpata DM, Blatnik JA, Novitsky YM, Rosen MJ
(2012) Modified hernia grading scale to stratify surgical site
occurrence after open ventral hernia repairs. J Am Coll Surg
215(6):787–793
3. Hodgkinson JD, Maeda Y, Leo CA, Warusavitarne J, Vaizey CJ
(2017) Complex abdominal wall reconstruction in the setting of
active infection and contamination: a systematic review of hernia
and fistula recurrence rates. Colorectal Dis 19:319–330
4. Dindo D, Demartines N, Clavien P-A (2004) Classification of
surgical complications: a new proposal with evaluation in a cohort
of 6336 patients and results of a survey. Ann Surg 240(2):205–213
5. Krpata D, Stein S, Eston M, Ermlich B, Blatnik JA, Novitsky Y
(2013) Outcomes of simultaneous large complex abdominal wall
reconstruction and enterocutaneous fistula takedown. Am J Surg
205:354–359
6. Sbitany H, Kwon E, Chern H, Finlayson E, Varma MG (2015)
Outcomes analysis of biologic mesh use for abdominal wall recon-
struction in clean-contaminated and contaminated ventral hernia
repair. Ann Plastic Surg 75:201–204
7. Rosen M, Krpata D, Ermlich B (2013) A 5-year clinical expe-
rience with single-staged repairs of infected and contaminated
abdominal wall defects utilizing biologic mesh. Ann Surg
257:991–996
8. Slater N, Bokkerink W, Konijn V, Bleichrodt R (2015) Safety
and durability of one-stage repair of abdominal wall defects with
enteric fistulas. Ann Surg 261:553–557
9. Berger RL, Li LT, Hicks SC, Davila JA, Kao LS, Liang MK
(2013) Development and validation of a risk-stratification score
for surgical site occurrence and surgical site infection after open
ventral hernia repair. J Am Coll Surg 217(6):974–982
10. Liang MK, Goodenough CJ, Martindale RG, Roth JS, Kao LS
(2015) External validation of the ventral hernia risk score for
prediction of surgical site infections. Surg Infect (Larchmt)
16(1):36–40
11. Vaizey CJ, Maeda Y, Barbosa E, Bozzetti F, Calvo J, Irtun Ø etal
(2016) ESCP consensus on the surgical management of intestinal
failure in adults. Color Dis 18:535–548
12. Haskins IN, Horne CM, Krpata DM, Prabhu AS, Tastaldi L, Perez
AJ etal (2018) A call for standardization of wound events report-
ing following ventral hernia repair. Hernia 22(5):729–736
13. Atema JJ, de Vries FEE, Boermeester MA (2016) Systematic review
and meta-analysis of the repair of potentially contaminated and con-
taminated abdominal wall defects. Am J Surg 212(5):982–995.e1
14. Atema JJ, Mirck B, Van Arum I, Ten Dam SM, Serlie MJ, Boer-
meester MA (2016) Outcome of acute intestinal failure. Br J Surg
103(6):701–708
15. de Vries FEE, Atema JJ, Lapid O, Obdeijn MC, Boermeester MA
(2017) Closed incision prophylactic negative pressure wound ther-
apy in patients undergoing major complex abdominal wall repair.
Hernia 21(4):583–589
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.