Background

Recognising dying is difficult. We believe there is a predictable biological process to dying and previously demonstrated that urinary volatile organic compounds change in the last weeks and days of life of patients with lung cancer. We further analysed our urine samples using a different metabolomic platform, Liquid Chromatography QTOF Mass Spectrometry (LC-QTOF-MS).

Method

We prospectively collected urine samples from people with lung cancer many of whom were in the last 4 weeks of life. The samples were analysed using a LC-QTOF-MS.

Volcano plots identified metabolites that changed 2 fold for different time periods (0-28 days, 0-14 days, 0-7days, 0-5 days and 0-3 days).

All metabolites were also grouped into weeks. A One-way ANOVA between the groups identified metabolites that changed significantly.

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

Results

234 urine samples from 112 patients were analysed by LC-QTOF-MS. 90 metabolites were identified that increase or decrease in the last weeks or days. Pathway Analysis using MetaboAnalyst demonstrated a number of biochemical pathways affected during different time intervals; 0-2 weeks and 0-3 days before death.

Cox LASSO regression analysis was performed for the last 90 days. A model using 19 metabolites, prognosticates for each day in the last 90 days with high AUC values (88-90%). Patients can be categorized into high, medium and low risk of death. A Kaplan-Meier survival analysis demonstrated the groups were well separated.

Conclusion

• The results confirm urine metabolites predict when people with lung cancer are in the last weeks and days of life.
• Our model prognosticates for each of the last 90 days of life and is approximately 88% -90% accurate.
  

Impact statement

• This is the only model able to prognosticate for the last week or days of life.