Inter-rater reliability of the Barthel Index

SIR—We have read with great interest the paper of Sainsbury et al. [1] focusing on the reliability of the Barthel Index (BI) when used with older people. We have recently evaluated the inter-rater reliability of two groups, one composed of clinical nurses expert in Gerontology and the other composed of Family Medicine residents when they were faced with BI [2]. The reliability of the BI has been well documented in stroke patients [1, 3] and it seems that it may be useful even when a single training session is offered to non-health care interviewers [4]. Nevertheless, we hypothesised that a larger training period may be necessary. One hundred community-dwelling patients were assessed during a hospital admission episode due to diverse conditions. Eighteen patients were excluded because they were able to answer only a few questions. Previously, two weeks before admission, the BI (scored 0–100) [5] was evaluated through an interview. Sequential BI was assessed by clinical nurses in Gerontology and by residents of Family Medicine who had received a brief training session on BI, and additional scoring guidelines were facilitated. Previously, in a study that had included 30 patients, an excellent inter-rater reliability was found between the two nurses in the total BI (Spearman; r=0.98) or in individual activities (Cohen’s kappa higher to 0.89 in all). When necessary (presence of cognitive impairment, delirium) the interviewers obtained responses from carers (n=21). The student’s t test, the chi-squared or Fisher’s exact test and the Spearman correlation test were used. Reliability was assessed with Cohen’s kappa test. The sample was composed of 45 women (55%) and 37 men, with a mean age of 72.6 (±11) years. Mean BI, assessed by nurses, was 87.3 (±19) and by residents, 88 (±21). In 40 (48.7%) patients some differences in the mean values of BI were observed. When overall BI score was analysed, agreement was high (r=0.793), but the agreement inter-observer was low (kappa<0.4) in some fields (Table 1) such as feeding, grooming and transfers, and was medium in others (kappa from 0.40 to 0.75). When differences were analysed between those patients showing concordance and those who did not show it, the latter were older (76.2 versus 69.2; P<0.005), but there were no differences related to gender (P=0.6), either when the interview was done with the patient or with the carer (P=0.06). Inter-rater reliability of the Barthel Index (Cohen’s kappa) results for individual activities Inter-rater reliability of the Barthel Index (Cohen’s kappa) results for individual activities We have observed that although the overall concordance was good, it was weak in some subsets. These results suggest that the comparison should be done item by item; those items that showed higher differences were feeding, grooming and transfer. In conclusion, a low inter-rater reliability in the BI was observed when clinical nurses in Gerontology and Family Medicine residents were compared, thus suggesting that these need specific training in order to improve their BI use.