Using the modified Barthel index to estimate survival in cancer patients in hospice: observational study

Patients, methods, and results

We studied two samples of patients with cancer from the same hospice to generate and test the model. We determined sample sizes empirically from patients admitted consecutively over two different periods of two months (January-February and March-April 1998), in whom the modified Barthel index was determined weekly from admission for the duration of inpatient stay. Barthel score at admission, mean weekly change in score during inpatient stay (defined as final score minus admission score divided by length of stay), and survival from date of admission were recorded.

The two populations were similar with respect to Barthel score at admission, length of stay, and survival (table). In sample 1, survival correlated with Barthel score at admission (r_s=0.25, P=0.014) but more closely with mean weekly change (r_s=-0.52, P<0.001). To examine this relation further, three groups were pragmatically constructed from the first sample on the basis of mean weekly change in Barthel scores. These represented clinical patterns commonly seen in terminally ill patients: stable physical functioning (no loss of points), moderate deterioration (1-9 points lost per week), and marked deterioration (10 or more points lost per week).

This model was applied to sample 2 to assess its ability to estimate survival. Survival correlated with Barthel score at admission (r_s=0.3, P=0.002) but more closely with mean weekly change (r_s=-0.52, P<0.001). Corresponding groups between samples had similar median survival, but the differences in survival between the three groups within each sample were significant (table).

Comment

In terminally ill patients in a hospice, rates of change were more important indicators of survival than absolute measures. Mean change in weekly Barthel scores was calculated to provide a crude clinical marker of changing physical function. Using mean change assumes that the modified Barthel index is an interval measure, but this has not been supported.4 Despite this, half of patients with advanced cancer who lose 10 or more points per week die within two weeks (95% confidence interval 8.6 days to 19.4 days), and three quarters are dead at three weeks. In contrast, 50% of patients in whom the weekly score does not deteriorate survive for two months (35.2 days to 76.8 days).

Although Barthel score at admission correlated with overall survival, no differences in scores on admission were found among the three groups in either sample (sample 1, P=0.08, and sample 2, P=0.74, Kruskal-Wallis; see table on website). Admission score therefore cannot be used to determine pattern of subsequent change and hence to estimate survival more accurately.