Descriptive statistics were generated. Participants were analysed for the absence (score = 0) or presence (score = 1) of significant clinical prediction rules variables at 4, 6, 8 and 12 months (see Figure 1, and the clinical prediction rules instructions in Appendix 2 in the eAddenda). Validity and cohort contamination effects of prosthetic use behaviours were compared by plotting pattern of non-use over time for the retrospective and prospective cohorts. The retrospective study’s continuous variable thresholds were used to generate dichotomous classification of these continuous variables in the present prospective

study. To validate the clinical prediction rules for each of the time frames, chi-square tests were calculated to generate a progressive list of likelihood ratios (negative and positive, 95% CI) to determine the cumulative effect of having a number (ie, 1, 2, 3 etc) of these Luminespib non-use predictors. Sensitivity, specificity, positive find more prediction value, accuracy and balanced accuracy were calculated to define

the accuracy and precision of clinical prediction rules in the prospective cohort.32 For both the retrospective and prospective statistical analyses, in circumstances where zero cases were present in frequency cells of the 2 x 2 contingency tables, 0.5 was added to the cell values to enable calculation of the likelihood ratios for the variables.33 Extreme likelihood ratio upper confidence limits were truncated at 999. Sensitivity analyses of 29 (16%) retrospective and eight (10%) prospective deceased prosthetic rehabilitation

participants who could not be interviewed were performed for 4, 6, 8 and 12 months after discharge to identify the presence or absence of clinical prediction below rules variables using date of death as the termination date for prosthetic use. Table 2 summarises the consecutive participants’ eligibility for the study. The final response rates were 94% (n = 135) for the retrospective cohort and 97% (n = 66) for the prospective cohort. The retrospective cohort were interviewed at median = 1.9 years (IQR 1.4 to 2.5) and prospective at median 1.3 years (IQR 1.1 to 1.4) after discharge. Table 3 outlines the geographical distribution of participants, as measured by Accessibility Remoteness Index of Australia.34 Clinical prediction rules development interviews with the retrospective cohort were performed by telephone (n = 123), telehealth (n = 2) and in person (n = 10). Twelve interviews were performed with carer assistance due to language interpretation, hearing or intellectual disability. Clinical prediction rules validation interviews with the prospective cohort were performed by telephone (n = 47) and in person (n = 19). Carers assisted with two interviews where participants had a hearing or intellectual disability. Table 3 shows the retrospective and prospective cohort characteristics.