A simple method to estimate survival trajectories Dr. Matt Williams ICHNT & IC E-Oncology Feb 2015 [email protected].

A simple method to estimate survival

trajectoriesDr. Matt Williams

ICHNT & IC

E-OncologyFeb 2015

[email protected]

The need

• 331 000 people diagnosed a year with cancer (UK, 2011)• 161 823 cancer deaths (UK, 2012)• Lung, bowel, breast, prostate – 54% cases, 46% of deaths

• May be diagnosed with incurable disease (e.g. Lung)• May initially be treated curatively, and then relapse (e.g Breast)

• Once diagnosed with “incurable” disease, may still be treated and live for many years

CRUK Cancer Stats (2015)

Talking about survival

• Curable/ incurable• “In the long run, we are all dead”

• Prognostic factors• Bias survival one way or another• Patient & Tumour factors (age, fitness; tumour grade, molecular aspects)• Clinicians are bad at predicting prognosis

• Survival remains a process, of a population, over time

JM Keynes, 1923

Looking at survival

Williams et al., Clin Onc, 2013

How can we characterise survival?

Stockler et al., BJC, 2006

If it is exponential

• Constant risk• E.g. radioactive decay

• Multiples of the median estimate other points• 75% ~ median/2• 25% ~ median x 2

• Assumes no cure• No conditional survival

Metastatic Colorectal cancer

• Trials of patients receiving first-line palliative chemotherapy• 2000 – 2011, phase III, 2+ regimens, 100+ pts per arm, 75% of pts had died

• 46 trials• 96 curves for analysis• 96 points at 90%, 75%, 25%; 54 points at 10%

• Obtained median survival• Median/4; median/2; median * 2; median *3

• Agreement defined calculated being 0.75 – 1.33 actual figure

Metastatic Colorectal cancer

• 46 trials; 29 011 patients• Median OS 16.8 (IQR: 14.3 – 19.4)• 342 data points• 301 (88%) acceptable• Worst agreement at 90% level (76% agreement)• Tendency to underestimate time to 90% and 75%, over-estimate to 25% and

10%

Williams et al., Ann Onc, 2014

Related work

• Breast, Lung and Prostate cancer• We now have data on cancer accounting for ~ half all cancer deaths• GBM in progress

• Clinicians aren’t accurate, but are good enough at estimating the median survival• We are discussing collaboration with Sydney group

Kiely et al., JCO 2013Kiely et al., JCO, 2011Kiely et al., Lung Cancer , 2012West et al., EJC, 2014

Computational aspects

• Very simple computation !• (1/4, ½, *2, *3)

• Based on a mathematical understanding of an empirical observation• Widely applicable• Helps us think about clinical practice

• Orthogonal to other prognostic tools• Better prognostic estimates improve estimates of the median

Thanks

• Anna Lerner & Ramsay Singer• Martin Utley (UCL) for discussion

• ICHNT & ICRUK centre supporting my work