A prospective study examining cachexia predictors in patients with incurable cancer

Abstract

Background: Early intervention against cachexia necessitates a predictive model. The aims of this study were to identify predictors of cachexia development and to create and evaluate accuracy of a predictive model based on these predictors.

Methods: A secondary analysis of a prospective, observational, multicentre study was conducted. Patients, who attended a palliative care programme, had incurable cancer and did not have cachexia at baseline, were amenable to the analysis. Cachexia was defined as weight loss (WL) > 5% (6 months) or WL > 2% and body mass index< 20 kg/m². Clinical and demographic markers were evaluated as possible predictors with Cox analysis. A classification and regression tree analysis was used to create a model based on optimal combinations and cut-offs of significant predictors for cachexia development, and accuracy was evaluated with a calibration plot, Harrell's c-statistic and receiver operating characteristic curve analysis.

Results: Six-hundred-twenty-eight patients were included in the analysis. Median age was 65 years (IQR 17), 359(57%) were female and median Karnofsky performance status was 70(IQR 10). Median follow-up was 109 days (IQR 108), and 159 (25%) patients developed cachexia. Initial WL, cancer type, appetite and chronic obstructive pulmonary disease were significant predictors (p ≤ 0.04). A five-level model was created with each level carrying an increasing risk of cachexia development. For Risk-level 1-patients (WL < 3%, breast or hematologic cancer and no or little appetite loss), median time to cachexia development was not reached, while Risk-level 5-patients (WL 3-5%) had a median time to cachexia development of 51 days. Accuracy of cachexia predictions at 3 months was 76%.

Conclusion: Important predictors of cachexia have been identified and used to construct a predictive model of cancer cachexia.

Trial registration: ClinicalTrials.gov Identifier: NCT01362816 .

Keywords: Cachexia; Cancer; Palliative care; Pre-cachexia; Weight loss.

Fig. 1

Flow chart

Fig. 2

Classification and regression tree (CART) analysis. The study population is divided repeatedly according to optimal cut-offs of the variables weight loss (rounded to the nearest integer), cancer type and appetite loss into subdivisions of significantly different hazard rates. Adjacent subdivisions from different branches with similar hazard rates are combined resulting in five risk-levels. Hazard ratios (HR) are reported relative to the branch with neutral risk cancer type and no or little appetite loss

Fig. 3

Kaplan-Meier plot of time to cachexia development depending on risk-level. Median time to cachexia development was not reached in level 1, 249 days for level 2, 175 days for level 3, 145 days for level 4 and 51 days for level 5. Log-rank test and test for trend in failure time-analysis were both significant (p <  0.0001)

Fig. 4

Calibration plot showing the risk of cachexia development after 3 months, as predicted by the risk-level model, plotted against the observed risk

Fig. 5

Sensitivity and specificity of cachexia prediction at 3 months when using different cut-offs of risk-level to divide patients into a high or low risk group of cachexia development. Risk-level ≥ 2 yields a high sensitivity (95%), while risk-level ≥ 3 yields a high specificity (88%). No single cut-off yields both a high sensitivity and high specificity