Oral Presentation Clinical Oncology Society of Australia Annual Scientific Meeting 2018

Understanding underuse of surgery for early non-small cell lung cancer – contribution of enhanced comorbidity ascertainment from population-level hospital and pharmaceutical dispensing data (#76)

Nicola Creighton 1 , Amy Johnston 1 , Hui You 1 , Laura Holliday 1 , David Currow 1
  1. Cancer Institute NSW, Alexandria, NSW, Australia


Underuse of curative treatment for early non-small lung cancer (NSCLC) has been reported nationally. Comorbidity can contraindicate surgery. However, population-based studies have generally relied on hospital-reported comorbidities which may not be comprehensively captured. This study measured comorbidity across hospital and pharmaceutical dispensing data and assessed the impact on use of surgery for early NSCLC in NSW.


Population-level linked data from the NSW Admitted Patient Data Collection and Pharmaceutical Benefits Scheme were obtained for NSW residents with a localised NSCLC diagnosed January 2009 - June 2012 on the NSW Cancer Registry using probabilistic privacy-preserving data linkage. Surgery with curative intent was identified from the hospital admission data. Measures of comorbidity were ascertained from hospital and drug dispensing data. Use of surgery was modelled using multivariable logistic regression.


Around half (52%, 673/1305) of people diagnosed with localised NSCLC did not undergo a major lung resection. Some comorbidities were captured better in hospital data compared to drug dispensing data and vice versa. E.g. dementia was ascertained for 1% v 7% and COPD for 45% v 23% of people from dispensing and hospital admission data respectively. Those who did not undergo surgery had higher medication use, particularly for COPD medications compared with those who received surgery. E.g. 31% of people who did not undergo surgery were dispensed a long-acting anticholinergic prior to diagnosis compared with 17% of those who had surgery (p<0.001). Including multiple measures of comorbidity and patient health ascertained across the study datasets increased the ability to predict resection status, with an increase in c-statistic from a logistic regression model from 0.85 using risk-factors typically available in population-based studies to 0.91.


Population-level drug dispensing data increases the ability to measure comorbidity which in turn enables a better understanding of observed gaps in optimal care for people with lung cancer.