Objective:

To adapt the Elixhauser comorbidity index by developing cancer-specific weights for seven cancer populations.

Methods:

This cohort study using SEER-Medicare linked data (2007-2016) included older patients (age>=65) diagnosed with cancer of the breast, lung, liver, prostate, pancreatic, kidney, or leukemia. Patients were included if they had continuous enrollment in Medicare part A and B with no Health Maintenance Organization enrollment in the year before cancer diagnosis. Following the approach of Mehta et al. (2018), each cancer cohort was split into 2 subsets (training and validation). Using the training cohort, Fine and Gray’s competing risk regression was used to model the association of Elixhauser comorbidities with time to non-cancer mortality within 2 years, with cancer mortality considered as a competing risk. Non-cancer mortality was the preferred outcome due to the emphasis on deaths attributable to comorbid conditions. Beta coefficients obtained were multiplied by 10 and rounded to the nearest integer to derive the cancer-specific weight for each comorbidity. Competing risk regression was used again in the validation cohort to compare the total Elixhauser score derived using individual comorbidities with the cancer-specific score while controlling for sex, age and cancer stage. Model performance was assessed by comparing c-statistics.

Results:

The final cohort comprised of breast (n=120,208), lung (n=107,129), liver (n=16,658), prostate (n=28,118), pancreatic (n=30,737), kidney (n=28,118), and leukemia (n=26,367). The 2-year non-cancer mortality rate was the highest for liver (14.07%) followed by leukemia (12.65%), lung (9.6%), kidney (8.68%), pancreatic (6.80%) and breast (4.8%) cancer. Cancer-specific Elixhauser score (c statistics: breast=0.55, lung=0.62, liver=0.61, pancreas=0.62, prostate=0.55, kidney=0.56 leukemia=0.57), performed slightly better than total Elixhauser score (c statistics: breast=0.55, lung=0.61, liver=0.59, prostate=0.55, pancreas=0.62, kidney=0.55, leukemia=0.57).

Conclusion:

Non-cancer mortality was the highest with liver cancer and the lowest with breast cancer patients. Cancer-specific Elixhauser score may control confounding better than the total Elixhauser score.