Material throughout this chapter has been reproduced from Cockayne et al.2 © 2021 Cockayne S et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Participant flow
Participants were mainly enrolled into OTIS via mail-outs from GP surgeries or from previous trial cohorts. The flow of participants is illustrated in the CONSORT flow diagrams in –. Across the eight participatory sites (East Coast Community Healthcare, East Sussex Healthcare NHS Trust, Harrogate and District NHS Foundation Trust, Humber Teaching NHS Foundation Trust, Leeds Community Healthcare NHS Trust, Northern Lincolnshire and Goole NHS Foundation Trust, Sheffield Teaching Hospitals NHS Foundation Trust and York Teaching Hospital NHS Foundation Trust), 12 ‘centres’ were formed. These centres were formed for logistical reasons to stratify the randomisation and were based on the geographical areas covered by the OTs.
The CONSORT flow diagram, up to randomisation, for GP and Yorkshire Health Study cohort mail-outs. a, More than one reason can apply.
The CONSORT flow diagram depicting the flow of participants from randomisation. a, Includes withdrawals from trial/death where this was before the intervention was received; b, withdrawals and deaths over time are cumulative.
The CONSORT flow diagram, up to randomisation, for YTU existing trial cohort mail-outs. a, More than one reason can apply. Reproduced with permission from Cockayne et al. © 2021 Cockayne S et al. This is an open access article distributed under (more...)
General practitioner mail-outs
A total of 11,965 recruitment packs were mailed or handed out to potential participants between March 2017 and April 2018 from GP surgeries, via opportunistic screening or through the University of the Third Age. The geographical locations covered included Harrogate, York and Elvington, Sheffield, Grimsby, East Coast Community (across Norfolk and Suffolk) and East Sussex.
Yorkshire Health Study
A recruitment pack was sent to 1314 participants from the Yorkshire Health Study cohort between April and July 2017.
Existing trial cohorts
A total of 6029 potentially eligible participants from previous trials conducted at YTU were mailed between October 2016 and March 2018: 3142 (52.1%) from CASPER (CollAborative care and active surveillance for Screen-Positive EldeRs),20 1741 (29%) from SCOOP (screening of older women for prevention of fracture)21 and 1146 (19.0%) from REFORM.19
Recruitment
Recruitment commenced in October 2016 and ceased when the final participant was randomised in August 2018. Overall, 19,308 recruitment packs were distributed. Among these, no response was received to 15,491 (80.2%), 162 were returned as ‘addressee unknown’, 159 people had died, seven participants received more than one pack (duplicates), three packs were returned too late to be included in the trial, and two participants were out of area. A further 384 people returned incomplete documentation. In total, 3100 (16.1%) potential participants returned a screening questionnaire and a valid consent form and were assessed for eligibility; 68 (2.2% of 3100) declined to participate, 1468 (47.4%) were immediately eligible and 1564 (50.4%) were initially ineligible. The most predominant reason for ineligibility was not having had a fall in the previous 12 months or not having a fear of falling (n = 1361, 87.0%), although this was usually not the only reason ().
Reasons for ineligibility
Based on their initial screening, 1289 participants were otherwise eligible except that they had not had a fall in the previous 12 months or did not have a fear of falling. These participants were eligible to be rescreened. A rescreening questionnaire was sent to 965 people (among the rest, they either declined to be rescreened or recruitment had closed before they were due to be rescreened). Of the 147 participants who returned a rescreening form, 53 (36.1%) subsequently became eligible (43 of whom went on to be randomised). Eligible and consenting participants were sent a baseline questionnaire and a pack of falls calendars (n = 1496). Twenty-five eligible, consenting participants were not sent a baseline pack because the trial had closed to recruitment (n = 8); the participant was at a duplicate address (n = 7); the participant withdrew consent (n = 5); the participant lived outside an area that an OT could visit (n = 3); an imminent OT visit was planned (n = 1); and the participant had died (n = 1). Of the 1410 participants (94.3% of 1496) who returned a baseline questionnaire, 1354 (96.0% of 1410) also returned at least one falls calendar. Of these 1354 participants, 1331 were randomised [the remaining 23 either withdrew (n = 21) or died (n = 2) before they were randomised].
The overall randomisation rate, from the total number of recruitment packs sent out, was 6.9%. The rate varied according to the mode of recruitment. From 6029 recruitment packs mailed out from YTU trial cohorts, 714 (11.8%) participants were randomised, relative to 59 out of 1314 (4.5%) from the Yorkshire Health Study and 558 out of 11,965 (4.7%) from GP surgeries.
Randomisation
The first participant was randomised on 31 January 2017 and the last was randomised on 2 August 2018 (), with follow-up ending in August 2019. Participants were randomised in 168 batches of between 2 and 32 patients. In total, 1331 participants were randomised into OTIS: 430 (32.3%) to the intervention group and 901 (67.7%) to the usual-care group (see ). We therefore exceeded our target of 1299 by 32 participants (), albeit with the requirement of an extension to the recruitment period from October 2017, by which time we initially had hoped to complete recruitment, to August 2018. A median of 66 participants were recruited from each of the 12 centres (range 19–312 participants).
Monthly recruitment to the OTIS trial.
Actual and target recruitment to the OTIS trial.
There was a median of 27 days (interquartile range 20 to 40 days) between completion of the screening questionnaire and completion of the baseline questionnaire. Participants were randomised a median of 44 days (interquartile range 25–73 days) after completing their baseline questionnaire. This allowed them time to return at least one falls calendar and for the OTs to confirm their capacity to deliver the intervention visits.
Baseline data
Baseline data for the 1331 randomised participants are presented in and . The mean age of participants was 80 years (range 65–98 years), and two-thirds (n = 872, 65.5%) were female. Three-quarters (n = 999) of the participants had sustained a fall in the 12 months prior to enrolment, among whom one in five (19.7%) had attended a hospital for treatment following their fall. The two groups were comparable on all baseline characteristics, except for a chance imbalance in the proportion of participants with Parkinson’s disease. Participants in the intervention group (n = 14, 3.3%) were more likely to have Parkinson’s disease than those in the usual-care group (n = 9, 1.0%).
Baseline demographic characteristics of randomised participants
Baseline falls and balance data for randomised participants
Comparing randomised participants with ineligible or non-consenting participants, non-randomised participants tended to be very slightly younger (mean age 78.5 years) and less likely to be female (58.0%).
Number of falls calendars returned
The response rates for the monthly falls calendars, where month 0 is the month of randomisation, are presented in . Overall, the response rate per month was consistently > 90%; however, the response rate is lower in the intervention group each month than in the usual-care group. This difference increases from 1.2% (97.2% compared with 98.4%) at month 0 to 3.4% (87.9% compared with 91.3%) at month 12.
Return rates of calendars post randomisation
Participant questionnaire return rates
Within 4 months of randomisation, there were two reported deaths (both in the usual-care group) and 28 withdrawals from the trial [16 (3.7%) in the intervention group and 12 (1.3%) in the usual-care group]. These participants were therefore not sent a 4-month participant questionnaire [1301 (97.8%) were sent] (). Between 4 and 8 months post randomisation, a further nine deaths (intervention group, n = 5; usual-care group, n = 4) and 24 withdrawals (intervention group, n = 9; usual-care group, n = 15) were reported. Between 8 and 12 months post randomisation, 15 deaths (intervention group, n = 7; usual-care group, n = 8) and 16 withdrawals (intervention group, n = 4; usual-care group, n = 12) were reported. Therefore, at 12 months, 389 (90.5%) participants randomised to the intervention group and 848 (94.1%) in the usual-care group were sent a follow-up questionnaire. Overall, participant response rates to the follow-up questionnaires at 4, 8 and 12 months were consistently above 90%. The response rates, with the number sent as the denominator, were similar between the two groups across all three time points; however, when using the number randomised as the denominator, the response rates decrease over time and are lower in the intervention group than in the usual-care group. At 12 months, 87.7% of the intervention group returned a questionnaire compared with 91.5% of the usual-care group. This reflects that the intervention group had a higher withdrawal rate and so a higher proportion of participants in this group were not sent the questionnaire.
Participant questionnaire response rates and time to response by randomised group at 4, 8 and 12 months post randomisation
Occupational therapist-delivered environmental assessment and modification visits
A total of 382 participants allocated to the intervention group received an environmental assessment and modification visit from an OT. Of these participants, 362 (94.8%) completed the Timed Up and Go test, with a mean of 15.6 seconds (SD 8.4 seconds, range 5 to 70 seconds); 159 (43.9%) scored over 14 seconds, indicating a high risk of falling. The assessments were conducted by 23 OTs (median 16 visits per OT, range 1 to 54 visits per OT). Nineteen of the OTs attended a face-to-face training session, and four were ‘cascade’ trained by another OT who had attended face-to-face training. The visits took place between 1 and 411 days after randomisation (median 27 days), and lasted a median of 90 minutes (range 25 to 180 minutes). Nearly two-thirds of the intervention group (277/430, 64.4%) had received the visit within 6 weeks of being randomised, and 381/430 (88.6%) received it within 12 months. The delays in delivering the visits were due to availability of the participant and, despite agreeing to the number of participants to be randomised at a given time, OT capacity. One participant received the visit beyond 12 months after they were randomised as they lived on the border of two trusts and it could not be agreed which trust should undertake the visit. A total of 48 participants did not receive a visit.
Of the 48 participants randomised to the intervention group who did not receive a visit, 25 withdrew from the intervention and 16 withdrew fully from the trial before receiving a visit. Reasons for withdrawing from the intervention were as follows: the participant did not feel that they would benefit from an OT visit (often as they felt that they were fit and well), n = 12; an OT visit was not appropriate (house was well equipped/had ongoing renovations/was a rental property/had been already assessed and adapted for needs of spouse), n = 4; the OT was unable to arrange visit, n = 3; the participant lived outside the area that the OT would attend, n = 3; no reason given, n = 2; and the participant did not want a visit due to ill health, n = 1. Reasons for full withdrawal among these participants were broadly similar, as it was common for participants to inform the OT when they came to arrange the appointment that they did not want the visit (i.e. they were withdrawing from treatment). These participants were then asked if they would be willing to continue to provide outcome data, which some declined to do; hence, they were fully withdrawn from the trial.
Seven participants appeared to remain in full participation with the trial but did not receive an environmental assessment. It is possible that these participants received a home visit but, as no documentation relating to the visit was received for them, we have conservatively assumed these participants did not receive the intervention.
Primary outcome
Raw data
In total, 1303 (97.9%) trial participants returned at least one falls calendar following randomisation (intervention, n = 419, 97.4%; usual care, n = 884, 98.1%), with 1204 (90.5%) returning a complete 12 months’ worth of calendars post randomisation (intervention, n = 377, 87.7%; usual care, n = 827, 91.8%). Of the 28 participants who did not return any falls calendars, five (four usual care, one intervention) provided falls data on at least one of the 4-, 8- and 12-month participant questionnaires. These data were used in the analysis for these participants. In total, 2260 falls were reported: 826 in the intervention group (mean 1.9 falls, SD 5.5 falls; median 1 fall, range 0–94 falls) over an average of 338 days (median 365 days), and 1434 in the usual-care group (mean 1.6 falls, SD 3.0 falls; median 1 fall, range 0–41 falls) over a mean of 345 days (median 365 days).
At least some information, such as the location and perceived cause, was available for 2037 (90.1%) falls (intervention, n = 700, 84.7%; usual care, n = 1337, 93.2%; ). Just over half of falls for which there was available location information occurred indoors (53.4%), with the majority (85.7%) of these occurring inside the participant’s own home rather than inside another premises. About half of the falls resulted in a superficial injury or worse, with 2.8% of the falls resulting in a broken bone (from 16 falls in the intervention group and 41 falls in the usual-care group), most frequently in the wrist or the hand (17/57, 29.8%).
Details of reported falls
Covariates
The primary analysis model controlled for sex, age at randomisation, history of falling and allocation ratio, and centre. Full covariate data were available for all randomised participants so the primary model was based on 1331 participants.
Primary analysis
The adjusted negative binomial model indicated weak evidence of a difference in falls, with an increase in the fall rate in the intervention group relative to usual care (IRR 1.17, 95% CI 0.99 to 1.38; p = 0.07). History of falling was seen to be a significant predictor in the model (IRR 3.15, 95% CI 2.69 to 3.69; p < 0.001).
Sensitivity analyses
Post hoc adjustment for Parkinson’s disease
One participant, in the intervention group, reported 94 falls over the 12 months from randomisation. This number of falls was significantly greater than the next largest of 41 (in the usual-care group). We observed that this participant had Parkinson’s disease, and also that there was a chance imbalance in the proportion of participants in the two groups who had Parkinson’s disease. We therefore decided, as a post hoc sensitivity analysis, to repeat the primary analysis, including whether or not the participant had Parkinson’s disease as an additional fixed effect. The resulting IRR was decreased to 1.11 (95% CI 0.94 to 1.31) and there was no evidence of a difference between the two groups (p = 0.23).
Non-compliance
When non-compliance with the intervention was accounted for using an instrumental variable CACE analysis approach, defining compliance as receipt of an OT-delivered environmental assessment within 12 months of randomisation, the CACE estimate of the intervention effect was very similar to that from the ITT analysis (IRR 1.18, 95% CI 0.98 to 1.43; p = 0.08).
Missing data
Ten (2.3%) participants in the intervention group and 13 (1.4%) participants in the usual-care group did not provide any falls data post randomisation. These participants were still included in the model as zero falls over a negligible time frame of 0.1 months was imputed for them. However, we conducted a logistic regression to assess whether or not any baseline factors were associated with missing outcome data and whether or not a participant reporting a fall in the 12 months prior to completing the screening questionnaire was the only statistically significant predictor. Participants who had had at least one previous fall were more likely to have missing outcome data (OR 5.84, 95% CI 1.13 to 30.21; p = 0.04). When this was added as a covariate in the primary analysis, the intervention effect estimate was virtually unchanged (IRR 0.17, 95% CI 0.99 to 1.38; p = 0.07).
Therapist effects
In three of the 12 centres, only one OT delivered the intervention visits. In the remaining eight centres, up to four OTs were involved. To account for potential therapist effects, we assigned each participant to an OT (i.e. the one they were seen by or a counterfactual therapist they could have seen had they been randomised to the intervention group). Therapist (rather than centre) was then included as a random effect in the primary analysis model. The intervention effect estimate was virtually unchanged (IRR 0.17, 95% CI 0.99 to 1.38; p = 0.07).
Subgroup analysis
About one-tenth (n = 124) of randomised participants (intervention group, 41/430, 9.5%; usual-care group, 83/901, 9.2%) reported that they had attended a hospital (outpatient appointment, day case, A&E presentation or overnight admission) for a fall in the 4 months prior to completing the baseline questionnaire.
We repeated the primary analysis in the subgroups of those who did and those who did not receive hospital treatment as a result of a fall in the 4 months prior to baseline and found qualitatively dissimilar treatment effects in each (received treatment: IRR 0.86, 95% CI 0.50 to 1.47; received no treatment: 1.21, 95% CI 1.01 to 1.44). However, when an interaction between this factor and treatment allocation was included in the primary model, the interaction was not observed to be statistically significant (p = 0.24).
Secondary analyses
Number of falls as reported for the previous 4 months on the 4-, 8- and 12-month participant questionnaires
In the intervention group, 404 participants (94.0%) provided a valid response to the number of falls they had experienced over the previous 4 months on at least one of the 4-, 8- and 12-month questionnaires. An average of 2.0 falls was reported (range 0–67 falls) over an average of 11.4 months. In the usual-care group, 870 participants (96.6%) provided a valid response to the number of falls they had experienced over the previous 4 months on at least one of the 4-, 8- and 12-month questionnaires. An average of 1.8 falls was reported (range 0–43 falls) over an average of 11.5 months. After imputing zero falls over a negligible time period of 0.1 months for those who did not provide any data, the adjusted IRR obtained from the negative regression model was 1.05 (95% CI 0.89 to 1.25; p = 0.56).
Comparison of the number of falls reported on falls calendars and in participant questionnaires
We compared the number of falls reported by participants in the questionnaires and on falls calendars. Of the 1331 randomised participants, 1132 (85.0%) returned 12 months’ worth of falls calendars and provided data on retrospective falls in all three participant questionnaires (4, 8 and 12 months). Among these, in the intervention group, the average number of falls reported on calendars was 1.9 (SD 5.7, median 1, range 0–94 falls) and in questionnaires was 2.0 (SD 5.0, median 1, range 0–67 falls). The correlation between the two counts was high, at 0.93 (95% CI 0.91 to 0.94). In the usual-care group, the average number of falls reported on calendars was 1.6 (SD 2.9, median 1, range 0–41 falls) and in questionnaires was 1.8 (SD 3.2, median 1, range 0–43), with correlation at 0.89 (95% CI 0.88 to 0.90). depicts a Bland–Altman plot of the agreement between falls reported on the calendars and falls reported in the questionnaires. There is strong agreement; the mean difference is –0.16 (participants tended to report marginally more falls in the questionnaires than on the calendars). The 95% limits of agreement are –3.5 to 3.2. There is a hint of a trend that the higher the average count of falls (between the two measurements), the larger the difference between the two tends to be.
Bland–Altman plot of the agreement of falls reported in questionnaires and on calendars.
Proportion of fallers and multiple fallers
In total, 245 out of 430 (57.0%) intervention group participants and 506 out of 901 (56.2%) usual-care group participants reported at least one fall on their monthly falls calendars or in participant questionnaires (adjusted OR 1.06, 95% CI 0.83 to 1.34; p = 0.65).
Using only data from the participant questionnaires, 230 out of 430 (53.5%) randomised participants in the intervention group and 506 out of 901 (56.2%) participants in the usual-care group reported at least one fall. There was no evidence of a difference in the likelihood of reporting a fall between the two groups (adjusted OR 0.91, 95% CI 0.72 to 1.15; p = 0.42).
These analyses, by default, assume that (1) participants who did not return any falls data did not fall and (2) partial responders had no falls in the months for which no data were provided. In terms of point 1, 57 (4.3%) participants failed to provide any falls data in the participant questionnaires, and only 23 (1.7%) participants failed to provide any falls data either on the monthly calendars or in the participant questionnaires. At the extremes, we could assume that all 57 participants with missing questionnaire falls data in each group, in turn, did fall at least once in the 12 months following randomisation. Assuming that the 26 participants in the intervention group fell (but leaving the usual-care group unchanged) the adjusted OR favours usual care, but the result is not statistically significant (1.17, 95% CI 0.92 to 1.49; p = 0.19). Conversely, assuming that the 31 participants in the usual-care group fell (but leaving the intervention group unchanged), there is weak evidence of a reduction in the proportion of participants in the intervention group who experienced at least one fall (adjusted OR 0.79, 95% CI 0.63 to 1.01; p = 0.06). The conclusion that there is no difference in the proportion of fallers in the two groups is, therefore, reasonably robust.
Proportion of multiple fallers (those having two or more falls in the 12 months from randomisation)
The proportion of participants who reported two or more falls on their falls calendars (or participant questionnaires, where no falls calendars were returned) following randomisation was also slightly higher in the intervention group than in the usual-care group (148/430, 34.4%, vs. 298/901, 33.1%; adjusted OR 1.11, 95% CI 0.86 to 1.43; p = 0.42).
Using only data from the participant questionnaires, 149 out of 430 (34.7%) randomised participants in the intervention group, and 323 out of 901 (35.9%) randomised participants in the usual-care group reported at least two falls. There was no evidence of a difference in the likelihood of reporting multiple falls between the two groups (adjusted OR 0.96, 95% CI 0.75 to 1.24; p = 0.78).
Similarly, these analyses must be interpreted with the following in mind. By default, they assume that (1) participants who did not return any falls data actually fell no more than once in the 12 months from randomisation, (2) partial responders who reported one fall experienced no falls in all the months for which no data were provided; and (3) partial responders who reported no falls experienced no more than one fall in all the months for which no data were provided. No further analyses to test the robustness of these results to these assumptions were deemed warranted, as such a large proportion of participants (90%) returned 12 months’ worth of falls calendars.
Proportion of participants having at least one fracture over the 12 months’ follow-up
A total of 54 participants reported a fracture from a fall in the 12 months from randomisation (intervention group, 16/430, 3.7%; usual-care group, 38/901, 4.2%).
Proportion of participants obtaining multiple fractures
Only two participants, both in the usual-care group, reported more than one fracture resulting from a fall, from separate events, in the 12 months from randomisation. One reported three fractures (wrist, ribs and knee) and the other reported two fractures (wrist and knee).
Proportion of participants who reported that they were worried about falling at 12 months post randomisation (reported worrying about falling at least some of the time)
Fear of falling was measured at screening and in the 4-, 8- and 12-month questionnaires using the question ‘During the past 4 weeks how often have you worried about having a fall?’, with the following response categories:
At 12 months, 197 out of 375 (52.5%) participants in the intervention group and 440 out of 818 (53.8%) participants in the usual-care group (where the denominator represents those providing a valid response to this question) reported that they worried about falling at least some of the time (some, a good bit, most or all of the time). There was no evidence of a difference in the likelihood of participants in the intervention group reporting a fear of falling relative to the usual-care group (adjusted OR 1.00, 95% CI 0.78 to 1.29; p = 1.00).
Fear of falling in its continuous form
Raw scores for fear of falling are summarised in and were fairly consistent between groups and over time. At screening, the average score was 4.3 (SD 1.3) in the intervention group and 4.2 (SD 1.3) in the usual-care group. At 12 months, the average score was 4.2 (SD 1.3) in the intervention group and 4.1 (SD 1.3) in the usual-care group. There was no evidence of a difference in fear of falling between the two groups at any post-randomisation time point (see ; ).
Unadjusted and adjusted results for analysis of fear of falling outcome in its continuous form
Adjusted means for fear of falling outcome by randomised group.
Time to fall
The median time to fall was 119 days in the intervention group (95% CI 105 to 133 days) and 144 days in the usual-care group (95% CI 132 to 155 days). Kaplan–Meier survival curves are presented for each group in . The adjusted hazard ratio from the Cox proportional hazards model for the intervention effect was 1.24 (95% CI 0.94 to 1.63; p = 0.12), indicating that the hazard or chance of falling at any particular time was higher in the intervention group than in the usual-care group, but this ratio is not statistically significant. This analysis assumes that no falls were experienced in the months for which no falls calendar was returned; no further investigation or consideration of imputation of data was deemed necessary to evaluate the impact of this assumption, as such a high proportion of randomised participants (90%) returned a full 12 months’ worth of falls calendars.
Kaplan–Meier curves by randomised group for time to fall.
Adverse events
There were no serious or non-serious adverse events that were related to the intervention or trial procedures and that needed to be reported in accordance with the protocol.
