Machine learning to predict early recurrence after oesophageal cancer surgery

BACKGROUND

Early cancer recurrence after oesophagectomy is a common problem, with an incidence of 20-30 per cent despite the widespread use of neoadjuvant treatment. Quantification of this risk is difficult and existing models perform poorly. This study aimed to develop a predictive model for early recurrence after surgery for oesophageal adenocarcinoma using a large multinational cohort and machine learning approaches.

METHODS

Consecutive patients who underwent oesophagectomy for adenocarcinoma and had neoadjuvant treatment in one Dutch and six UK oesophagogastric units were analysed. Using clinical characteristics and postoperative histopathology, models were generated using elastic net regression (ELR) and the machine learning methods random forest (RF) and extreme gradient boosting (XGB). Finally, a combined (ensemble) model of these was generated. The relative importance of factors to outcome was calculated as a percentage contribution to the model.

RESULTS

A total of 812 patients were included. The recurrence rate at less than 1 year was 29·1 per cent. All of the models demonstrated good discrimination. Internally validated areas under the receiver operating characteristic (ROC) curve (AUCs) were similar, with the ensemble model performing best (AUC 0·791 for ELR, 0·801 for RF, 0·804 for XGB, 0·805 for ensemble). Performance was similar when internal-external validation was used (validation across sites, AUC 0·804 for ensemble). In the final model, the most important variables were number of positive lymph nodes (25·7 per cent) and lymphovascular invasion (16·9 per cent).

CONCLUSION

The model derived using machine learning approaches and an international data set provided excellent performance in quantifying the risk of early recurrence after surgery, and will be useful in prognostication for clinicians and patients.

Multicentre cohort study to define and validate pathological assessment of response to neoadjuvant therapy in oesophagogastric adenocarcinoma

BACKGROUND

This multicentre cohort study sought to define a robust pathological indicator of clinically meaningful response to neoadjuvant chemotherapy in oesophageal adenocarcinoma.

METHODS

A questionnaire was distributed to 11 UK upper gastrointestinal cancer centres to determine the use of assessment of response to neoadjuvant chemotherapy. Records of consecutive patients undergoing oesophagogastric resection at seven centres between January 2000 and December 2013 were reviewed. Pathological response to neoadjuvant chemotherapy was assessed using the Mandard Tumour Regression Grade (TRG) and lymph node downstaging.

RESULTS

TRG (8 of 11 centres) was the most widely used system to assess response to neoadjuvant chemotherapy, but there was discordance on how it was used in practice. Of 1392 patients, 1293 had TRG assessment; data were available for clinical and pathological nodal status (cN and pN) in 981 patients, and TRG, cN and pN in 885. There was a significant difference in survival between responders (TRG 1-2; median overall survival (OS) not reached) and non-responders (TRG 3-5; median OS 2·22 (95 per cent c.i. 1·94 to 2·51) years; P < 0·001); the hazard ratio was 2·46 (95 per cent c.i. 1·22 to 4·95; P = 0·012). Among local non-responders, the presence of lymph node downstaging was associated with significantly improved OS compared with that of patients without lymph node downstaging (median OS not reached versus 1·92 (1·68 to 2·16) years; P < 0·001).

CONCLUSION

A clinically meaningful local response to neoadjuvant chemotherapy was restricted to the small minority of patients (14·8 per cent) with TRG 1-2. Among local non-responders, a subset of patients (21·3 per cent) derived benefit from neoadjuvant chemotherapy by lymph node downstaging and their survival mirrored that of local responders.

Impact of Food and Predation on the Snowshoe Hare Cycle

Snowshoe hare populations in the boreal forests of North America go through 10-year cycles. Supplemental food and mammalian predator abundance were manipulated in a factorial design on 1-square-kilometer areas for 8 years in the Yukon. Two blocks of forest were fertilized to test for nutrient effects. Predator exclosure doubled and food addition tripled hare density during the cyclic peak and decline. Predator exclosure combined with food addition increased density 11-fold. Added nutrients increased plant growth but not hare density. Food and predation together had a more than additive effect, which suggests that a three-trophic-level interaction generates hare cycles.

Specificity of association between Bacillus isolates and genotypes of Lolium perenne and Trifolium repens from a grass – legume pasture

Specificity between plants and associated rhizosphere bacteria was investigated using species and genotypic mixtures of Lolium perenne L. (perennial ryegrass) and Trifolium repens L. (white clover) inoculated with Bacillus strains isolated from rhizosphere soil of the pasture plants. The genotypic identity of plants was controlled by using stolon tips (Trifolium) and tillers (Lolium) of three genotypes of each species collected from a 45-year-old permanent pasture. Inoculation of plants in the greenhouse with Bacillus isolates that had coexisted in the field with the Lolium component of a Lolium–Trifolium mixture increased Lolium root and shoot weight. Root and nodule weight of Trifolium in mixture regardless of its genotype was also higher when coexistent Bacillus–Lolium combinations were present. The presence of other coexistent Bacillus–plant or plant–plant combinations did not enhance performance of either pasture species. The identity of the Rhizobium leguminosarum biovar trifolii strains that formed Trifolium root nodules was unaffected by inoculation with Bacillus. Key words: Trifolium repens, Lolium perenne, Bacillus, specificity, rhizosphere, yield.

The influence of Rhizobium leguminosarum biovar. trifolii on the growth and neighbour relationships of Trifolium repens and three grasses

Trifolium repens was sampled from 10 areas dominated by each of Dactylis glomerata, Holcus lanatus, and Lolium perenne in each of three different-aged pastures. Cell extracts of nodule isolates of Rhizobium leguminosarum biovar. trifolii were compared using polyacrylamide gel electrophoresis. Each pasture contained a range of Rhizobium strains as defined by the protein profiles. No single strain was common to Trifolium sampled from the same grass species neighbourhood. The diversity of Rhizobium strains encountered was negatively correlated (r = −0.67; n = 9) with pasture age; old pastures showed a decrease in strain diversity. In addition, in the oldest pasture (45-year-old), tillers of the associated grasses were also sampled. In a glasshouse, ramets from each Trifolium clone and each of the three grass species were grown in all possible combinations with each of the three Rhizobium strains, plus Trifolium monocultures and an uninoculated control. All Trifolium clones gave increased yield in mixture with Lolium compared with yields with Dactylis and Holcus. Trifolium from the Holcus patches, in most cases, had higher yield than Trifolium from other grass patches. Rhizobium inoculation did not consistently increase Trifolium yield; it actually caused a decrease in yield when in mixture with Lolium. A significant Rhizobium × Trifolium interaction indicated a variable performance of each Trifolium genotype with different Rhizobium strains. The identity of the Trifolium clone had a significant effect on the yield of Holcus and Lolium, but not Dactylis. Total mixture yields were highest when Trifolium from a Lolium patch was part of the mixture. These results are discussed in the context of the hypothesis that Trifolium–grass coexistence is significantly influenced by Rhizobium strain differences.