Background Recognising dying is difficult and is an ongoing difficulty for doctors. We believe there is a process to dying and animal studies support this theory. We previously demonstrated that a number of volatile organic compounds in urine, change in the last weeks and days of life of patients in a small mixed cancer group. We needed to verify this finding in a suitably powered follow-up study.

Method We prospectively collected urine samples from people with lung cancer. We aimed to compare samples from 25 people in each of the last 3 weeks of life to a control group, 50 people with lung cancer who lived 3 or more months from the time of sampling. The urine samples were analysed for volatile organic compounds by gas chromatography mass spectrometry (GC-MS).

Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression was used to analyse the GC-MS data and create a statistical model.

Results We recruited 162 people in total; 29 in the last week; 28 in the second last week; 30 in the third last week of life; 74 controls i.e. samples taken >3 months from death; 424 urine samples.

A model was created to predict whether a patient would die within 1 week. It has an optimism corrected AUC of 0.851 (95% CI: 0.767, 0.911); sensitivity 78.6% (95% CI: (64.3%, 89.3%)); specificity 83.1% (95% CI: (69.9%, 92.3%)). The model identified a selection of compounds that contributed to the identification of patients who were close to death.

Discussion

• The results confirm that volatile organic compounds can predict when people with lung cancer are in the last week of life.

• Our model to predict when a person with lung cancer is in the last week of life is approximately 80% accurate.