Roger Vaughan

Interpretable machine learning has been focusing on explaining final models that optimize performance. The current state-of-the-art is the Shapley additive explanations (SHAP) that locally explains variable impact on individual predictions, and it is recently extended for a global assessment across the dataset. Recently, Dong and Rudin proposed to extend the investigation to models from the same class as the final model that are "good enough", and identified a previous overclaim of variable importance based on a single model. However, this method does not directly integrate with existing Shapley-based interpretations. We close this gap by proposing a Shapley variable importance cloud that pools information across good models to avoid biased assessments in SHAP analyses of final models, and communicate the findings via novel visualizations. We demonstrate the additional insights gain compared to conventional explanations and Dong and Rudin's method using criminal justice and electronic medical records data.

Primary Mycobacterial infection of the sternum is an exceedingly rare occurrence. We present a case of tuberculous osteomyelitis of the sternum successfully treated with surgical debridement and vacuum-assisted closure therapy.

We investigated the pathological outcome of lung resections undertaken for solitary pulmonary nodules (SPNs) <30 mm in diameter in a regional thor

Foreign body ingestion is a common problem in children. Most ingested foreign bodies pass through the gastrointestinal tract without difficulty once they reach the stomach. Perforation of the gastrointestinal tract very rarely occurs. This case report discusses the clinical presentation and management of a very unusual case of an eyeliner pencil ingested by a child that perforated the stomach and diaphragm, causing empyema.

BACKGROUND: The significance of allogeneic blood transfusion in the prognosis of patients with nonsmall lung carcinoma (NSCLC) remains controversial. Reports have suggested that patients with adenocarcinoma have a worse prognosis from cancer than patients with squamous cell carcinoma (SCC), but this evidence is lacking in NSCLC. The objective of the current study was to elucidate the correlation between perioperative allogeneic blood transfusion and the prognosis in patients with adenocarcinoma and SCC.METHODS: The study group comprised 329 consecutive patients (172 men and 157 women) with a mean age of 67 years who underwent lung resection between 1996 and 2003 in 1 unit. The clinicopathologic and survival data were compared between 62 patients (42.7%) with adenocarcinoma and 58 patients (48.3%) with SCC who received a perioperative blood transfusion and 83 patients (57.3 %) with adenocarcinoma and 126 patients (61.3%) with SCC who did not.RESULTS: The overall surgical mortality rate was 3.9% (13 deaths). The median surgical blood loss was 380 mL (range, 125–4500 mL). The mean blood transfusion received was 1.9 U (range, 0–8 U). The overall actuarial 1, 3, and 5‐year survival rates in the adenocarcinoma and SCC groups were 82.3%, 49.6%, and 19.4% and 87.9%, 71.7%, and 32.7%, respectively (P = 0.021). Multivariate analysis demonstrated that the factors that appeared to be independent prognostic factors in both groups were T‐classification (P < 0.001), lymph node metastasis (P < 0.001), and postoperative complications (P = 0.029). Perioperative blood transfusion was not an independent prognostic indicator in either group.CONCLUSIONS: The current study demonstrated that long‐term survival in patients with adenocarcinoma of the lung was significantly worse compared with patients with SCC, but was independent of allogeneic perioperative blood transfusion. The results reaffirmed the importance of tumor invasion and lymph node involvement in the overall poor prognosis of these patients. Cancer 2004. © 2004 American Cancer Society.

Objective: Cervical mediastinoscopy is an important diagnostic and staging technique. Limited operative field and visibility have traditionally made it a difficult procedure to learn and supervise. Video-assisted techniques can aid training in the procedure. We designed a prospective study to assess the usefulness of video-assisted mediastinoscopy (VAM) as a training tool. Methods: 43 patients were operated upon by two trainees during their initial formation in general thoracic surgery (25 patients in 15 months, and 18 patients in 9 months, respectively). Indications: staging (n=23), diagnosis of enlarged mediastinal nodes (n=14), and diagnosis/staging (n=6). End-points of the study: operative time, need of consultant assistance during procedures, and ability of the trainee to identify all nodal stations independently. Results: There were no complications. The mean operative time was 29 (range 18–51) min. Valid histological samples were obtained in all cases. There were no false negative results in the 13 patients who underwent subsequent lung resection (sensitivity 100%). Operative time (R²=0.83 and 0.77), need for consultant assistance (R²=0.98 and 0.94), and failure to independently reach all nodal stations (R²=0.95 and 0.94) significantly decreased with experience in both trainees' cases (cubic curve fit; PDiscussion: VAM permits a rapid learning and adequate supervision of the technique without compromising safety, operative time or completeness of the procedure. The main advantages are: increased visual field, image magnification, adequate light source and the ability to use two instruments simultaneously. VAM should be the technique of choice in thoracic surgical teaching units.

We describe the thoracoscopic management of mediastinal sepsis and foreign body associated with a cervical oesophageal perforation. Following initially successful conservative management, the patient's condition worsened and we undertook surgical exploration where the abscess was drained and a fragment of china recovered.

We prospectively audited the implementation of a protocol for management of air-leaks and intercostal drains after thoracic surgical procedures. Out of the 99 patients who underwent thoracic surgical procedures during a 3-month period, 41 presented a postoperative air-leak on day 1. The protocol was strictly followed in 95% of the cases. The median drainage-time and hospital-stay were 3 and 4 days, respectively. The incidence of insertion of postoperative drains was 3%. Most of the nursing staff found the protocol user-friendly and easy to apply. A protocol for the management of postoperative air-leaks and intercostal drains can be implemented with high-compliance low complications.