The predictive value of unexpected weight loss in adults for cancer in primary care: a matched cohort study using UK electronic health records data from 2000-2014.
Problem
A recent systematic review linked unexpected weight loss (UWL) with cancer diagnosis across ten sites in primary care. Important biases limited the meta-analysis of the 25 included studies. No study compared diagnostic intervals following the date of UWL or the predictive value of weight loss stratified by cancer site, cancer stage, gender and age-group.
Approach
We present a matched cohort analysis of representative electronic health records data from the UK’s CPRD for 2000-2014 with linkage to the National Cancer Registry. All patients aged >18 years old with an UWL code were matched 1:5 based on age, gender and GP practice to consulting patients without a weight loss code. Demographic, symptom and test result data were extracted in relation to the index date. Diagnostic intervals over a two year period were calculated and time-to-event modelling used to derive adjusted Hazard Ratios (HRs) over-time and by time-period, for clinically important subgroups.
Findings
Of 330,444 patients remaining after exclusions, 63,973 (19%) had UWL, 190,002 (57%) were females, and 124,953 (38%) were aged >70 years. The median (IQR) number of days to diagnosis for patients with and without UWL was 80 (26-290) vs. 353 (181-541), with marked variation by cancer site and gender. The adjusted 3-month and 6-month HRs (95% CI) for cancer were 3.3 (2.9-3.7) and 1.9 (1.7-2.1) in males and 1.9 (1.6-2.1) and 1.1 (0.99-1.3) in females. Late-stage cancer was more strongly associated with UWL than early stage. UWL combined with 30 individual clinical features and 20 abnormal test results pushed cumulative Positive Predictive Value for cancer above the 3% threshold for further investigation adopted by NICE.
Consequences
Our results will inform the optimal investigative strategy to diagnose cancer in patients presenting to primary care with UWL. To reduce unnecessary investigation we will also present the most powerful test combinations to rule-out the need for further investigation, informed by the cachexia literature.