Predicting rare outcomes in abdominal wall reconstruction using image-based deep learning models

Factors Associated With Respiratory Failure After Open Ventral Hernia Repair: An Evaluation of the NSQIP Database

An analysis of ACS-NSQIP open ventral hernia repair (OVHR) data (2017-2019) was performed. Respiratory failure (RF) occurred in 643 patients (1%) and not in 63,213 (99%) (nRF). Respiratory failure patients were older (63.7 vs 57 years, P < .001) and more comorbid: insulin-dependent diabetes (14.7% vs 5.8%, P < .001), COPD (19.4% vs 5.2%, P < .001), BMI (36.0 vs 32.8, P < .001), and current tobacco use (24.9% vs 17.6%, P < .001). Respiratory failure patients had greater ASA scores (ASA 3: 63.3% vs 47.8%, P < .001), bowel resection (8.2% vs 1.3%, P < .001), component separation (20.1% vs 9.0%, P < .001), operative times (178.4 vs 98.8 minutes, P < .001), complications (deep wound infections 3.6% vs 1.0%, organ space infections 13.2% vs 1.0%, wound dehiscence 3.1% vs 0.6%, acute renal failure 11.7% vs 0.1%), and hospital stay (13.7 vs 2.3 days), with fewer home discharges (44.3% vs 96.4%) (all P < .001). Respiratory failure patients had higher mortality compared to nRF (20.2% vs 0.1%, P < .001). Respiratory failure after OVHR is rare but correlates closely with significant wound, systemic, and social complications. Preoperative management of risk factors would be appropriate in high-risk patients.

Predicting response to neoadjuvant chemotherapy for colorectal liver metastasis using deep learning on prechemotherapy cross-sectional imaging

BACKGROUND AND OBJECTIVES

Deep learning models (DLMs) are applied across domains of health sciences to generate meaningful predictions. DLMs make use of neural networks to generate predictions from discrete data inputs. This study employs DLM on prechemotherapy cross-sectional imaging to predict patients' response to neoadjuvant chemotherapy.

METHODS

Adult patients with colorectal liver metastasis who underwent surgery after neoadjuvant chemotherapy were included. A DLM was trained on computed tomography images using attention-based multiple-instance learning. A logistic regression model incorporating clinical parameters of the Fong clinical risk score was used for comparison. Both model performances were benchmarked against the Response Evaluation Criteria in Solid Tumors criteria. A receiver operating curve was created and resulting area under the curve (AUC) was determined.

RESULTS

Ninety-five patients were included, with 33,619 images available for study inclusion. Ninety-five percent of patients underwent 5-fluorouracil-based chemotherapy with oxaliplatin and/or irinotecan. Sixty percent of the patients were categorized as chemotherapy responders (30% reduction in tumor diameter). The DLM had an AUC of 0.77. The AUC for the clinical model was 0.41.

CONCLUSIONS

Image-based DLM for prediction of response to neoadjuvant chemotherapy in patients with colorectal cancer liver metastases was superior to a clinical-based model. These results demonstrate potential to identify nonresponders to chemotherapy and guide select patients toward earlier curative resection.

Deep learning model utilizing clinical data alone outperforms image-based model for hernia recurrence following abdominal wall reconstruction with long-term follow up

BACKGROUND

Deep learning models (DLMs) using preoperative computed tomography (CT) imaging have shown promise in predicting outcomes following abdominal wall reconstruction (AWR), including component separation, wound complications, and pulmonary failure. This study aimed to apply these methods in predicting hernia recurrence and to evaluate if incorporating additional clinical data would improve the DLM's predictive ability.

METHODS

Patients were identified from a prospectively maintained single-institution database. Those who underwent AWR with available preoperative CTs were included, and those with < 18 months of follow up were excluded. Patients were separated into a training (80%) set and a testing (20%) set. A DLM was trained on the images only, and another DLM was trained on demographics only: age, sex, BMI, diabetes, and history of tobacco use. A mixed-value DLM incorporated data from both. The DLMs were evaluated by the area under the curve (AUC) in predicting recurrence.

RESULTS

The models evaluated data from 190 AWR patients with a 14.7% recurrence rate after an average follow up of more than 7 years (mean ± SD: 86 ± 39 months; median [Q1, Q3]: 85.4 [56.1, 113.1]). Patients had a mean age of 57.5 ± 12.3 years and were majority (65.8%) female with a BMI of 34.2 ± 7.9 kg/m2. There were 28.9% with diabetes and 16.8% with a history of tobacco use. The AUCs for the imaging DLM, clinical DLM, and combined DLM were 0.500, 0.667, and 0.604, respectively.

CONCLUSIONS

The clinical-only DLM outperformed both the image-only DLM and the mixed-value DLM in predicting recurrence. While all three models were poorly predictive of recurrence, the clinical-only DLM was the most predictive. These findings may indicate that imaging characteristics are not as useful for predicting recurrence as they have been for other AWR outcomes. Further research should focus on understanding the imaging characteristics that are identified by these DLMs and expanding the demographic information incorporated in the clinical-only DLM to further enhance the predictive ability of this model.

Machine learning, deep learning and hernia surgery. Are we pushing the limits of abdominal core health? A qualitative systematic review

INTRODUCTION

This systematic review aims to evaluate the use of machine learning and artificial intelligence in hernia surgery.

METHODS

The PRISMA guidelines were followed throughout this systematic review. The ROBINS-I and Rob 2 tools were used to perform qualitative assessment of all studies included in this review. Recommendations were then summarized for the following pre-defined key items: protocol, research question, search strategy, study eligibility, data extraction, study design, risk of bias, publication bias, and statistical analysis.

RESULTS

A total of 13 articles were ultimately included for this review, describing the use of machine learning and deep learning for hernia surgery. All studies were published from 2020 to 2023. Articles varied regarding the population studied, type of machine learning or Deep Learning Model (DLM) used, and hernia type. Of the thirteen included studies, all included either inguinal, ventral, or incisional hernias. Four studies evaluated recognition of surgical steps during inguinal hernia repair videos. Two studies predicted outcomes using image-based DMLs. Seven studies developed and validated deep learning algorithms to predict outcomes and identify factors associated with postoperative complications.

CONCLUSION

The use of ML for abdominal wall reconstruction has been shown to be a promising tool for predicting outcomes and identifying factors that could lead to postoperative complications.

Defining surgical risk in octogenarians undergoing paraesophageal hernia repair

BACKGROUND

With an aging population, the utility of surgery in elderly patients, particularly octogenarians, is of increasing interest. The goal of this study was to analyze outcomes of octogenarians versus non-octogenarians undergoing paraesophageal hernia repair (PEHR).

METHODS

The Nationwide Readmission Database was queried for patients > 18 years old who underwent PEHR from 2016 to 2018. Exclusion criteria included a diagnosis of gastrointestinal malignancy or a concurrent bariatric procedure. Patients ≥ 80 were compared to those 18-79 years old using standard statistical methods, and subgroup analyses of elective and non-elective PEHRs were performed.

RESULTS

From 2016 to 2018, 46,450 patients were identified with 5425 (11.7%) octogenarians and 41,025 (88.3%) non-octogenarians. Octogenarians were more likely to have a non-elective operation (46.3% vs 18.2%, p < 0.001), and those undergoing non-elective PEHR had a higher mortality (5.5% vs 1.2%, p < 0.001). Outcomes were improved with elective PEHR, but octogenarians still had higher mortality (1.3% vs 0.2%, p < 0.001), longer LOS (3[2, 5] vs 2[1, 3] days, p < 0.001), and higher readmission rates within 30 days (11.1% vs 6.5%, p < 0.001) compared to non-octogenarian elective patients. Multivariable logistic regression showed that being an octogenarian was not independently predictive of mortality (odds ratio (OR) 1.373[95% confidence interval 0.962-1.959], p = 0.081), but a non-elective operation was (OR 3.180[2.492-4.057], p < 0.001). Being an octogenarian was a risk factor for readmission within 30 days (OR 1.512[1.348-1.697], p < 0.001).

CONCLUSIONS

Octogenarians represented a substantial proportion of patients undergoing PEHR and were more likely to undergo a non-elective operation. Being an octogenarian was not an independent predictor of perioperative mortality, but a non-elective operation was. Octogenarians' morbidity and mortality was reduced in elective procedures but was still higher than non-octogenarians. Elective PEHR in octogenarians is reasonable but should involve a thorough risk-benefit analysis.