Machine learning algorithms to predict atypical metastasis of colorectal cancer patients after surgical resection

Colorectal cancer (CRC) is the second most common cause of cancer death in the United States. Every 3 years, the American Cancer Society provides an update of CRC occurrence based on incidence data (available through 2016) from population-based cancer registries and mortality data (through 2017) from the National Center for Health Statistics. In 2020, approximately 147,950 individuals will be diagnosed with CRC and 53,200 will die from the disease, including 17,930 cases and 3,640 deaths in individuals aged younger than 50 years. The incidence rate during 2012 through 2016 ranged from 30 (per 100,000 persons) in Asian/Pacific Islanders to 45.7 in blacks and 89 in Alaska Natives. Rapid declines in incidence among screening-aged individuals during the 2000s continued during 2011 through 2016 in those aged 65 years and older (by 3.3% annually) but reversed in those aged 50 to 64 years, among whom rates increased by 1% annually. Among individuals aged younger than 50 years, the incidence rate increased by approximately 2% annually for tumors in the proximal and distal colon, as well as the rectum, driven by trends in non-Hispanic whites. CRC death rates during 2008 through 2017 declined by 3% annually in individuals aged 65 years and older and by 0.6% annually in individuals aged 50 to 64 years while increasing by 1.3% annually in those aged younger than 50 years. Mortality declines among individuals aged 50 years and older were steepest among blacks, who also had the only decreasing trend among those aged younger than 50 years, and excluded American Indians/Alaska Natives, among whom rates remained stable. Progress against CRC can be accelerated by increasing access to guideline-recommended screening and high-quality treatment, particularly among Alaska Natives, and elucidating causes for rising incidence in young and middle-aged adults.

The prognostic nutritional index (PNI), which is calculated based on the serum albumin concentration and peripheral blood lymphocyte count, is a useful tool for predicting short-term and long-term postoperative outcome in patients undergoing cancer surgery. However, few studies have investigated PNI in colorectal cancer surgery. We examined the ability of PNI to predict short- and long-term outcomes in patients with colorectal cancer. This retrospective study included 365 patients who underwent resection for colorectal cancer. The prognostic nutritional status was calculated on the basis of admission data as follows: 10 × serum albumin (g/dl) + 0.005 × total lymphocyte count (per mm(3)). The primary outcomes measured were the impact of PNI on overall survival and postoperative complications. Kaplan-Meier analysis and the log rank test revealed that low PNI was significantly associated with poor survival (P < 0.0001). In multivariate analysis for survival, preoperative low PNI was an independent prognostic factor for poor survival: odds ratio: 2.25, 95 % confidence interval 1.42-3.59). Moreover, low PNI significantly correlated with the incidence of postoperative complications, especially serious ones. Preoperative PNI is a useful predictor of postoperative complications and survival in patients with colorectal cancer.

This study investigated whether nutritional risk scores applied at hospital admission predict mortality and complications after colorectal cancer surgery. Some 186 patients were studied prospectively. Clinical details, Reilly's Nutrition Risk Score (NRS) and Nutritional Risk Screening 2002 (NRS-2002) score, tumour stage and surgical procedure were recorded. The prevalence of patients at nutritional risk was 31.7 per cent according to Reilly's NRS and 39.3 per cent based on the NRS-2002. Such patients had a higher mortality rate than those not at risk according to Reilly's NRS (8 versus 1.6 per cent; P = 0.033), but not the NRS-2002 (7 versus 1.8 per cent; P = 0.085). Based on the NRS-2002, there was a significant difference in postoperative complication rate between patients at nutritional risk and those not at risk (62 versus 39.8 per cent; P = 0.004) but not if Reilly's NRS was used (58 versus 44.1 per cent; P = 0.086). Nutritional risk was identified as an independent predictor of postoperative complications (odds ratio 2.79; P = 0.002). Nutritional risk screening may be able to predict mortality and morbidity after surgery for colorectal cancer. However, the diverse results reflect either the imprecision of the tests or the small sample size. Copyright 2010 British Journal of Surgery Society Ltd. Published by John Wiley & Sons, Ltd.