Clinical effectiveness and cost-effectiveness of a multifaceted podiatry intervention for falls prevention in older people: a multicentre cohort randomised controlled trial (the REducing Falls with ORthoses and a Multifaceted podiatry intervention trial)

Exclusion criteria for the REFORM cohort

Participants were ineligible for the REFORM cohort if they:

1. were > 65 years of age

2. reported having neuropathy, dementia or another neurological condition such as Parkinson’s disease, Alzheimer’s disease, multiple sclerosis, Lou Gehrig’s disease/amyotrophic lateral sclerosis or Huntington’s disease

3. were unable to walk household distances (10 metres) without the help of a walking aid, such as a walking frame, a walker or a person to assist

4. had had a lower limb amputation

5. were unwilling to attend their podiatry clinic for a REFORM appointment.

Inclusion criteria for the REFORM cohort

All eligible consenting participants who completed a baseline questionnaire and at least one monthly falls calendar were eligible for inclusion in the REFORM cohort.

Inclusion criteria for the REFORM trial

Participants in the cohort were eligible for inclusion in the REFORM trial if they:

1. had had a fall in the past 12 months, or a fall in the past 24 months requiring hospital attention, or reported worrying about falling at least some of the time in the 4 weeks prior to completing their baseline questionnaire

2. were community dwelling

3. were able to read and speak English.

If participants did not report a recent fall on their screening form but later reported a fall on the baseline questionnaire or monthly falls calendar, they became eligible to be randomised.

Pilot sample size

The pilot phase of the study ran from October 2012 until November 2013. No formal sample size calculation was conducted but we aimed to randomise at least 70 participants into the pilot trial (35 in each group), which we believed to be sufficient to test the objectives.

REFORM trial sample size

The primary outcome measure for the trial was the incidence rate of falls reported by the participants over the 12 months post randomisation. This was analysed using a mixed-effects negative binomial regression model. However, because of the inherent difficulties of estimating the parameters required to power a trial for a count outcome, such as the IRR to detect and the measure of overdispersion, the trial was instead powered for the binary outcome of whether or not the participants experienced at least one fall, which was one of our key secondary outcomes. We retained incidence rate of falls as the primary outcome as we believed the extra information contained in this outcome would result in the sample size being conservative for this outcome.

A previous falls prevention trial conducted by some of the authors in community-dwelling older adults with a history of recent falls found a 12% absolute reduction in the proportion of participants who fell among those allocated to receive an environmental falls prevention intervention delivered by qualified occupational therapists, relative to the control group. 25 The REFORM trial was powered at 80% using a two-sided 5% significance level to detect a more conservative absolute difference of 10 percentage points from 50% to 40% in the number of people experiencing at least one fall over the 12 months following randomisation. The total sample size required, allowing for a 10% loss to follow-up, was 890 participants (445 in each group).

Intervention group

Participants in the intervention group were allocated to receive a multifaceted intervention comprising footwear advice (and footwear provision if required), an orthotic insole or review of an existing prescription, a programme of foot and ankle balance exercises and a falls prevention leaflet. The trial protocol recommended that participants be invited to attend two appointments: the first as soon as possible after randomisation and the second 2–4 weeks later. Further trial visits could be offered if required, in addition to routine podiatry care appointments in accordance with usual practice.

Primary outcome

The primary end point for the trial was the incidence rate of falls per participant in the 12 months following randomisation. A fall was defined as ‘an unexpected event in which the participant comes to rest on the ground, floor, or lower level’. 30 Data were collected via participant self-reported monthly falls calendars. These took the form of A5 pieces of card with a calendar grid of individual months printed on one side along with a definition of a fall and a freepost address to the YTU on the other. Participants were asked to record the day of the month on which they fell or to record that they did not fall that month and return the calendar to the YTU. Participants who did not return their monthly falls calendar were either telephoned or written to by the YTU to obtain the missing data. Participants were also given a freephone number to report any falls as soon as possible after they occurred, and these were recorded by research staff on a falls telephone data collection sheet (see Appendix 9). The information collected included the date and location of fall, the reason for the fall, any injuries sustained (e.g. a superficial wound or a broken bone), hospital admissions, the footwear worn at the time of the fall and if the participant was wearing an insole or using a walking aid.

Secondary outcomes

All secondary outcomes were self-reported by the participant and collected by questionnaires at 6 and 12 months post randomisation or by monthly falls calendars. Secondary outcomes include:

1. proportion of fallers (at least one fall and multiple falls)

2. time to first fall from date of randomisation

3. incidence rate of falls in 12 months post randomisation as recorded on the 6- and 12-month participant questionnaires

4. fear of falling as measured by the question ‘During the past 4 weeks have you worried about having a fall?’ at 6 and 12 months

5. fear of falling as measured by the Short Falls Efficacy Scale – International (FES-I) at 6 and 12 months31

6. activities of daily living as measured by the Frenchay Activities Index (FAI) at 6 and 12 months32

7. depression as measured by the short form Geriatric Depression Scale (GDS) at 6 and 12 months33

8. proportion of participants with depression (score of ≥ 6 on GDS) at 12 months

9. adaptability and resilience as measured by the 2-item abbreviated version of the Connor-Davidson Resilience Scale (CD-RISC2) at 6 months34

10. level of pain in the feet as measured on a visual analogue scale from 0 (no pain) to 10 (worst possible pain) at 12 months

11. fracture rate (single and multiple)

12. health-related quality of life (HRQoL) as measured by the EuroQoL-5 Dimensions-3 Levels (EQ-5D-3L)35

13. health service utilisation.

Fear of falling

Fear of falling was measured by the question ‘During the past 4 weeks have you worried about having a fall?’ at screening and baseline and at 6 and 12 months. Response categories were all of the time, most of the time, a good bit of the time, some of the time, a little of the time and none of the time. These were scored from 1 to 6 and were treated as continuous data for analysis.

Short Falls Efficacy Scale – International

The Short FES-I asked participants at baseline and at 6 and 12 months to indicate how concerned they were about falling when performing seven different activities: not at all concerned, somewhat concerned, fairly concerned or very concerned. These were scored from 1 to 4, and a total score for the Short FES-I was obtained by summing the seven item scores. When a participant selected two or more responses to an item, this was treated as missing data. If data were missing on two or more items then the questionnaire was considered invalid. If data were missing on no more than one of the seven items then the total score was calculated for the six completed items, divided by six and multiplied by seven. The new total score was then rounded up to the nearest whole number. 31 A final total score of 7 or 8 indicates no/low concern, 9–13 indicates moderate concern and 14–28 indicates a high degree of concern about falling.

Frenchay Activities Index

This 15-item instrument was administered at baseline and at 6 and 12 months and assessed a broad range of activities of daily living. The frequency with which each item or activity was undertaken over the previous 3 or 6 months (depending on the nature of the activity) was assigned a score of 1–4, where a score of 1 is indicative of the lowest level of activity (e.g. never performed). The scale provides a summed total score from 15 to 60. When a participant selected two or more responses to an item, this was treated as missing data. Only when there were no missing item responses was a total score computed for an individual for this instrument.

Geriatric Depression Scale

The GDS is a 15-item scale used as a screening tool for geriatric depression and was administered at baseline and at 6 and 12 months. Each item requires a ‘yes’ or ‘no’ response. A score of 1 is assigned when the item response indicates a negative state of mind, for example responding ‘no’ to ‘Are you basically satisfied with your life?’. A total score out of 15 can be calculated. When a participant selected both ‘yes’ and ‘no’, the worst-case scenario was assumed and a score of 1 was assigned to the item. More than five missing item responses invalidated the scale; otherwise, a total score when there were missing data was calculated by summing the item scores, dividing by the total number of completed items and multiplying by 15. The new total score was then rounded up to the nearest whole number to give the score for an individual (https://web.stanford.edu/∼yesavage/GDS.html). A score of 0–5 is considered normal, whereas a score of > 5 suggests depression. Any participant reporting a score of ≥ 10 on the GDS,33,36 that is, more severe depression, was referred to their GP.

The two-item abbreviated version of the Connor-Davidson Resilience Scale

The CD-RISC2 is a two-item abbreviated version of the full 25-item Connor-Davidson Resilience Scale. It is based on items 1 (‘I am able to adapt to change’) and 8 (‘I tend to bounce back after illness or hardship’) of the original instrument. Each item is scored from 0 (‘not true at all’) to 4 (‘true nearly all the time’), so the CD-RISC2 can be scored from 0 to 8. Higher scores reflect greater ‘bounce-back’ and adaptability. This instrument was administered at baseline and at 6 months.

Non-consenting participants

Participants who did not wish to take part in the study were not required to return any forms to the YTU; however, some chose to complete the screening form, thus providing us with some demographic information. In addition, all participants in the pilot phase of the study were sent an invitation pack that included a decline form, so that if they were willing they could provide a reason for declining. This provided us with sufficient information to document the reasons why participants did not wish to take part in the study, and allowed us to compare participants who declined with those who participated. The recruitment pack in the main study did not contain a decline form.

Intervention: details and adherence

Treatment details were recorded by the podiatrist, including the number of podiatry visits, an eligibility checklist with details on relevant health conditions and test results, characteristics of current indoor and outdoor shoes, details relating to shoes ordered, details on the type and prescription of any current insole use, the type of insole issued/retained with any modifications made, details of the size of any therapy ball and the strength of any resistive band prescribed and any amendments or advice given on the intervention owing to safety reasons.

Information on adherence to the exercise, footwear advice and orthotic insole components of the intervention was collected from participant self-reported questionnaires at 3, 6 and 12 months from participants in the intervention group only. Participants were asked if, during the past month, they had worn their insole all of the time, most of the time, some of the time, a little of the time or none of the time. Participants were also asked, for the past month, typically how many times a week they had done the exercises: not at all, once, twice, three times or more than three times. In addition, all participants were asked on the 12-month follow-up questionnaire if they had been given footwear advice by the trial podiatrist and whether or not they had followed the advice given.

Primary analysis

The primary analysis model controlled, as fixed effects, for sex (coded 0 = female, 1 = male), age at randomisation in years (integer) and history of falling. All participants had to have fallen at least once in the previous 12 months, have had a fall in the last 24 months requiring hospitalisation or have a fear of falling in order to be eligible for randomisation. Participants were classified into two groups for the history of falling covariate: (1) one or no falls in the 12 months prior to completion of the background information sheet; or (2) two or more falls reported in the 12 months prior to completion of the background information sheet. These were coded as 0 and 1, respectively.

As there was evidence of overdispersion in the data, Poisson regression was not considered to be appropriate, and so the incidence rate of falls was analysed using a mixed-effects negative binomial regression model. Participants recruited from the same centre, and therefore residing in a particular geographical area, are more likely to be similar to one another than to participants from other centres. This can result in a correlation between participant outcomes within centres. Failure to account for this clustering of outcomes in the analysis can lead to an increase in the type 1 error rate. Therefore, to account for the potential correlation of participant outcomes from participants in the same centre, we included trial centre (n = 13) as a random effect in the model. The model also took account of the different observation periods for each individual by including a variable for the number of months for which the participant returned a monthly falls calendar (using the exposure option within the Stata command).

The model equation is:

where

• E(y_ij) is the expected number of falls for participant i in centre j in time t_ij

• t_ij is the length of exposure (follow-up) for participant i in centre j

• β is a vector of fixed-effect regression coefficients and

• exp(β) is a vector of the IRRs.

Coefficients are presented as IRRs with 95% CIs and p-values.

Sensitivity analyses

A sensitivity analysis of the primary outcome was conducted, adjusting for any pre-randomisation variables found to be imbalanced by chance between the randomised groups.

Secondary analyses

The incidence rate of falls over the 12 months following randomisation (as reported for the previous 6 months on the 6- and 12-month participant questionnaires) was analysed in the same way as the primary outcome.

The proportion of fallers versus non-fallers, and of multiple fallers versus single or non-fallers, in each group was compared using a mixed logistic regression model adjusting for sex, age and history of falling, with centre included as a random effect. 37

The time from randomisation to first fall in days was derived. Participants who did not have a fall were censored at their date of death or, if alive, their withdrawal from the trial, the date of the last available assessment or 365 days after randomisation, whichever was latest. Kaplan–Meier survival curves were produced for each group. The time to first fall was analysed by a Cox proportional hazard regression with shared centre frailty effects adjusting for sex, age and history of falling. 38

Fear of falling in the past 4 weeks, and the total scores for the Short Falls Efficacy Scale – International, GDS and FAI were compared between the two groups using a covariance pattern mixed model incorporating all post-randomisation time points (6 and 12 months) adjusting for baseline score, sex, age, history of falling, treatment group, time and a treatment group-by-time interaction term, with centre as a random effect. Such an approach models the correlation of observations within participants over time. Different covariance structures for the repeated measurements, which are available as part of Stata version 13 (unstructured, exchangeable, independent and banded), were explored and the most appropriate pattern used for the final model based on the Akaike’s information criterion (smaller values are preferred). 39 Participants were included in the model if they had full data for the baseline covariates and outcome data for at least one post-randomisation time point (6 or 12 months). An estimate of the difference between treatment groups in the outcome was extracted for each time point with a 95% CI and p-value.

The assumptions of the covariance pattern mixed model were checked visually. The normality of the standardised residuals was assessed via a histogram and Q–Q plot, and the homoscedasticity of the errors was checked by plotting the residuals against the fitted values.

The CD-RISC2 score at 6 months was compared between the two groups using a linear mixed model adjusting for baseline CD-RISC2 score, sex, age and history of falling, with centre as a random effect.

Participants with a score of ≥ 6 on the GDS were categorised as having depression; the proportion of people with depression in each group was compared at 12 months using a mixed logistic regression model adjusting for sex, age and history of falling, with centre as a random effect.

The proportion of participants obtaining at least one fracture over the 12-month follow-up period was compared using a mixed logistic regression adjusting for sex, age and history of falling, with centre as a random effect.

At 12 months, participants were asked to indicate their level of pain or discomfort in their feet on a visual analogue scale from 0 (no pain) to 10 (worst possible pain). This was analysed using a linear mixed model adjusting for sex, age and history of falling, with centre as a random effect in an ITT analysis, and also in a CACE analysis. We based compliance on whether or not the participant was seen in clinic for a trial appointment. The CACE analysis employed a two-stage regression process: first, compliance with the intervention was predicted using a linear mixed model adjusting for randomised group allocation, sex, age and history of falling, with centre as a random effect; and second, foot pain score was predicted using a linear mixed model adjusting for compliance, sex, age, history of falling and the predicted residuals from the first regression, with centre as a random effect.

Design

A semistructured interview study was used to gather in-depth information on the trial participants’ experiences of receiving the podiatry intervention, alongside the podiatrists’ experiences of delivering the intervention. The interviews were conducted either face to face or over the telephone with participants and podiatrists in the trial (at the end of the intervention period).

Sampling

A purposive sampling strategy was used to achieve a hetergeneous sample of trial participants from the intervention group to ensure maximum variation40 according to age, sex and history of falls. Previous studies have indicated that a sample of approximately 20–30 trial participants is sufficient to address the aforementioned aims from the point of view of the service users.

As podiatrists delivering the intervention were based in a wide variety of clinics, it was expected that their views and experiences may differ. For example, some podiatrists worked in biomechanics and others worked in routine podiatry clinics. All 28 podiatrists who delivered the REFORM intervention were invited for interviews through the PI at each site.

Recruitment and consent

All REFORM trial participants living in the Yorkshire or Lincolnshire areas who expressed an interest in undertaking other associated REFORM research studies on the consent form and who had received the intervention were eligible for participation in the qualitative study. Following sampling, study participants were approached by letter, which contained an information sheet (see Appendix 15) and two consent forms (see Appendix 16). The letter also informed trial participants that a qualitative researcher (authors AC and SC) would contact them via telephone to find out if they would be willing to take part and, if so, to arrange a time for the interview to take place. In accordance with ethics guidelines, informed consent was gained by the researcher before the commencement of the interview. The aim of the interview was explained to the participant, and this was followed by an opportunity for them to ask questions about the study. The anonymity and confidentiality of participants’ personal information were assured by the researcher.

Podiatrists were also invited to take part in the qualitative interviews. The PI at each site was sent an e-mail asking if he or she and the podiatrists who delivered the intervention would like to be interviewed. The PI was asked to forward the e-mail on to podiatrists at their site who delivered the intervention. Podiatrists were asked to contact the research team directly if they wished to take part. The recruitment e-mail included an invitation, information sheet (see Appendix 17) and consent form (see Appendix 18). This was followed up by a telephone call or an e-mail. Prior to the interviews, podiatrists were assured anonymity and confidentiality and were given the opportunity to ask questions. For podiatrists interviewed face to face, a similar process to that used to obtain consent for trial participants was used. For interviews conducted over the telephone, verbal consent was obtained prior to the start of the interview and a copy of the consent form was sent to the qualitative researcher either in the post or via e-mail.

Data collection

The semistructured interviews with trial participants were carried out in participants’ homes or at the University of York between November 2013 and March 2016 and on average lasted 40 minutes using a topic guide (see Appendix 19). All interviews were audio-recorded, transcribed and anonymised before data analysis.

The semistructured interviews with podiatrists were carried out between July 2015 and January 2016 in a private room on premises where the podiatrist was based or over the telephone. The interviews lasted between 30 and 70 minutes and were conducted using a topic guide (see Appendix 20).

The topic guides provided a framework for the semistructured interviews and ensured that all podiatrists and trial participants were asked the same questions, allowing comparisons to be made during the analysis. However, the wording of questions was not fixed to allow interviews to flow and to allow for probing when more detail was required.

Data analysis

An initial thematic analysis was carried out using the stages as outlined by Braun and Clarke:41 (1) familiarisation, (2) generating initial codes, (3) searching for themes, (4) reviewing themes, (5) defining and naming themes and (6) data reporting. An initial coding framework was developed based on a priori themes relating to issues included in the topic guide while allowing for emergent themes. Descriptive coding was conducted, following familiarisation with the data, by the main qualitative researcher on the project (author SC), informed by regular discussion with the qualitative team (authors JA and AC). Subsequently, initial codes were refined in order to address the aims of the qualitative study outlined above, following the analysis of the main trial. A constant comparison method42 was used to check and compare across the data set and to establish appropriate analytical categories. This also ensured that any additional codes were added to reflect as many of the nuances or outlier views in the data as possible, taking into consideration the participants’ wider contexts. Anonymised participant identifiers are used for the reporting of results.

To promote quality, the following strategies were used: description of the participants to provide context (credibility and transferability), transparency of the research process (transferability), evidence of consistency using multiple examples from data (dependability) and engagement of the wider research team with interim findings (confirmability). In addition, a reflexive approach was taken to data analysis. The main interviewers (authors SC and AC) were academic research fellows with no podiatry training. SC was the main REFORM trial co-ordinator and AC had no prior knowledge or experience of podiatry interventions, orthotic insoles or RCTs. The other member of the qualitative team (author JA) had an academic research background and also did not have previous knowledge or experience of podiatry care. This placed the qualitative research team in a very neutral position relating to any prior expectations relating to the study intervention.

Pilot trial

During the pilot phase of the study, the following quantitative progression criteria were imposed to permit continuation to the main trial: (1) recruit 580 participants to the REFORM cohort study and (2) randomise 70 participants to the REFORM pilot trial. By the end of November 2013, 972 participants had been recruited to the REFORM cohort, and 78 had been randomised (39 participants per group) from York and Scarborough podiatry clinics.

Package of podiatry care

A baseline appointment form completed by the treating podiatrist was received for 380 participants. From this, we can summarise details of the intervention delivered, when this was provided. Twenty of 351 participants (5.7%) were reported as currently wearing custom-made shoes. Of the 364 participants who had their usual outdoor footwear assessed by the podiatrist at this visit, 249 (68.4%) were deemed to be wearing appropriate footwear; therefore, 115 (31.6%) exhibited a feature that made the shoes a risk for falling. The most common reason was inappropriate fixation/fastening (n = 81, 70.4%), followed by an inappropriate heel counter (n = 57, 49.5%), incorrect size (n = 43, 37.4%), unsuitable sole (n = 42, 36.5%), inappropriate heel width (n = 17, 14.8%) and excessive heel height (n = 15, 13.0%). New footwear was provided to 260 (52.7% of 493, 63.0% of 413) intervention participants throughout the course of the trial.

At the baseline appointment, 115 out of 363 (31.7%) participants were currently wearing an insole. The type of current insole being used was not recorded for six participants, and two participants reported wearing two different types of insole in different shoes or for different occasions. The most common type of insole being used was a simple flat-bed insole (n = 37, 32.2%), followed by a contoured prefabricated insole with simple modification (n = 25, 21.7%), a bespoke total contact insole (n = 23, 20.0%), a simple contoured prefabricated insole (n = 21, 18.2%) and a contoured prefabricated insole with complex modifications (n = 5, 4.4%).

An orthotic insole was fitted for 241 of the 413 (58.4%) participants who received the intervention (X-Line blue, n = 209; X-Line red, n = 23; Formthotics, n = 9). A therapy ball and theraband were prescribed for 355 (93.4%) and 358 (94.2%) participants, respectively, of the 380 for whom we have this information.

Raw data

In total, 992 (98.2%) trial participants returned at least one falls calendar following randomisation [intervention, n = 484, 98.2%; usual care, n = 508, 98.3%], with 762 (75.5%) returning a complete 12 months’ worth of calendars post randomisation (intervention, n = 360, 73.0%; usual care, n = 402, 77.8%). In total, 1423 falls were reported: 661 in the intervention group (median 1 fall, range 0–23 falls) over a median of 365 days (range 6–365 days), and 762 in the usual-care group (median 1 fall, range 0–28 falls) over a median of 365 days (range 27–365 days) (Figures 3 and 4).

FIGURE 3.

Histogram displaying the distribution of the number of falls by treatment group. (a) Intervention; and (b) usual care.

FIGURE 4.

Histogram displaying the distribution of the length of follow-up for falls data by treatment group. (a) Intervention; and (b) usual care.

Information, such as the cause and location, was available for 1172 (82.7%) falls (intervention, n = 549, 83.3%; usual care, n = 623, 82.1%; Table 17). Over one-third of the falls were reportedly caused by a trip (n = 457, 39%), and an injury was sustained in over half of the falls (n = 655, 55.9%). These injuries include 31 broken bones (from 17 falls in the intervention group and 14 in usual care). The most common bones broken in a fall were the hip or bones in the hand (n = 5 each).

Covariates

The primary analysis model controlled for sex, age at randomisation and history of falling. The age of one participant in the usual-care group was not available so the primary model was based on 991 participants (484 in the intervention group and 507 in the usual-care group).

Screening and baseline data for participants analysed in the primary model

Data collected on the screening form and baseline questionnaire are presented by randomised group for the 991 participants included in the primary analysis (‘as analysed’ population; Tables 18–21). The composition of the analysis groups is virtually identical to that at randomisation, indicating that the loss of the 19 participants from the primary analysis has not introduced any selection bias.

Number of falls by centre

A summary of the number of falls for participants contributing to the primary analysis is presented by centre in Table 22, in order of largest to smallest contributing centre.

Primary analysis

The adjusted negative binomial model indicated a non-statistically significant reduction in the fall rate in the intervention group relative to usual care (IRR 0.88, 95% CI 0.73 to 1.05; p = 0.16). History of falling was seen to be a significant predictor in the model (IRR 2.10, 95% CI 1.74 to 2.54; p < 0.001; Table 23). Little difference in the estimate of the treatment effect was observed when centre was included as a fixed, as opposed to a random, effect in the primary analysis model in a sensitivity analysis (IRR 0.88, 95% CI 0.73 to 1.07; p = 0.20).

Non-compliance

When non-compliance with the intervention was accounted for using an instrumental variable CACE analysis approach, the intervention was seen to have a marginally greater benefit than in the ITT analysis but the conclusions were otherwise consistent (IRR 0.86, 95% CI 0.69 to 1.06; p = 0.16).

Podiatrist effects

A single podiatrist appeared to deliver the intervention to all participants in 6 of the 13 centres. In the other seven centres, two, three, four (two centres each) or six (one centre) podiatrists held intervention appointments. Counterfactual podiatrists were assigned to the 80 participants in the intervention group who did not receive the intervention and the 517 usual care participants. Repeating the primary analysis with podiatrist as a random effect in the place of centre had a negligible effect on the treatment estimate (IRR 0.88, 95% CI 0.73 to 1.05; p = 0.16).

Baseline imbalance by chance

A sensitivity analysis of the primary outcome was planned, which adjusted the model for any pre-randomisation variables found to be imbalanced by chance between the randomised groups, namely proportion of participants wearing an insole, number of falls in 6 months prior to completion of baseline questionnaire and total FAI score. However, owing to concerns that this model could be overparameterised by including two variables relating to past falls history (a dichotomous variable indicating two or more falls in the 12 months prior to screening and a continuous variable of the number of falls recalled in the 6 months before baseline), the number of falls variable was not included.

The resultant model was based on 912 participants (intervention, n = 448, 90.9%; usual care, n = 464, 89.7%). Of those included in the primary model, 77 did not have a valid baseline FAI score and three did not provide a response to whether or not they were wearing an orthotic insole at screening; 79 participants were missing at least one of these additional covariates.

When these variables were added to the primary model, the predicted IRR was 0.88 (95% CI 0.72 to 1.06; p = 0.18), which is virtually unchanged from the primary model.

Fear of falling participants

Excluding the 26 fear of falling participants from the primary model had a negligible effect on the treatment estimate (IRR 0.88, 95% CI 0.73 to 1.05; p = 0.16).

Missing data

Not living with a partner or spouse, not having a degree or equivalent professional qualification, reporting dizziness or vertigo at screening and number of falls in the 6 months prior to completion of the baseline questionnaire were observed to predict returning less than 6 months’ worth of falls calendar data post randomisation. When these variables were included in the primary analysis model (excluding the number of falls variable for the same reasons as cited above), the parameter estimate for the treatment effect was IRR 0.86 (95% CI 0.72 to 1.04; p = 0.11). This model was based on 976 participants (intervention, n = 477, 96.8%; usual care, n = 499, 96.5%).

Number of falls as reported for the previous 6 months on the 6- and 12-month participant questionnaires

This outcome was computed for 450 (91.3%) participants in the intervention group and 484 (93.6%) in the usual-care group. In the intervention group, 423 (85.8%) participants responded to this question at both 6 and 12 months and 27 (5.5%) responded at either 6 or 12 months only; the average number of falls reported in this group was 1.5 (median 1 fall, range 0–20 falls). In the usual-care group, 457 (88.4%) participants responded to this question at both 6 and 12 months and 27 (5.2%) responded at either 6 or 12 months only; the average number of falls reported in this group was 1.7 (median 1 fall, range 0–29 falls). The adjusted IRR obtained from the negative regression model was 0.87 (95% CI 0.72 to 1.06; p = 0.17).

Proportion of fallers and multiple fallers

In total, 245 out of 493 (49.7%) intervention participants and 284 out of 517 (54.9%) usual care participants reported at least one fall on their monthly falls calendars (adjusted OR 0.78, 95% CI 0.60 to 1.00; p = 0.05). An OR of 0.6 (lower confidence limit) approximately relates to a decrease in the percentage of fallers from 55% in the usual-care group to 42% in the intervention group, which exceeds the 10 percentage point difference for which the trial was powered.

To calculate the percentage of the intervention group, let us say that the 2 × 2 table for falls by allocation is as given in Table 24.

In this case, we know that a + b = 493, c = 284, d = 233, c + d = 517 and N = 1010. The calculation for the OR is:

We want to know the values of a and b if c and d are as observed in the trial and the OR is 0.6. Rearranging (2), we get:

We know that a + b = 493, so if we rearrange and solve these equations, we conclude that a = 208 and b = 285. Therefore, if there were 208 fallers, this would equate to a percentage of 208/493 × 100 = 42%.

The analysis of fallers assumes that the 18 participants who did not return any falls calendars following randomisation into the trial did not fall. To test the sensitivity of the results to this assumption we repeated the logistic regression (1) dropping these participants and (2) assuming, conversely, that they did fall at least once. Estimates were robust: adjusted OR 0.77 (95% CI 0.59 to 0.99; p = 0.04); and OR 0.77 (95% CI 0.60 to 0.99; p = 0.05), respectively. In addition, if participants did not report a fall, it is implicitly assumed that these participants did not fall in the months for which they did not return a falls calendar. To test this, we assumed that a random 50% of the 185 participants [99/493 (20.1%) participants in the intervention group and 86/517 (16.6%) participants in the usual-care group] who did not report a fall and did not complete a falls calendar for all 12 months post randomisation fell at least once in a month for which data were missing. This increased the effect (adjusted OR 0.72, 95% CI 0.55 to 0.93; p = 0.01).

The proportion of participants who reported two or more falls on their falls calendars following randomisation was also lower in the intervention group than in the usual-care group [27.6% (n = 136/493) vs. 34.6% (n = 179/517); adjusted OR 0.69, 95% CI 0.52 to 0.90; p = 0.01].

Time to first fall

The median time to the first fall and its associated 95% CIs were estimated at 314 days (95% CI 267 days, upper limit not calculable) in the intervention group and 257 days in the usual-care group (95% CI 209 to 319 days). Kaplan–Meier survival curves are presented for each group in Figure 5. The adjusted hazard ratio from the Cox proportional hazards model for the treatment effect was 0.88 (95% CI 0.74 to 1.04; p = 0.13), indicating that the hazard or chance of falling at any particular time is lower in the intervention group than in the usual-care group, but this ratio is not statistically significant. Log-log plots of the categorical covariates indicated slight violation of the proportional hazard assumption for sex and allocation (and indeed we observe that the survival curves cross one another on the Kaplan–Meier curve but this occurs relatively early on); however, the Grambsch and Therneau test did not provide evidence that the assumptions did not hold.

FIGURE 5.

Kaplan–Meier curves by randomised group for time to first fall. Reproduced from Cockayne et al. 44 This is an open access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits unrestricted use, distribution and reproduction in any medium, provided the original author and source are credited (https://creativecommons.org/licenses/by/4.0/).

Participant-reported outcome measures

No statistically significant differences between the two groups were observed at 6 or 12 months in the fear of falling question, the Short FES-I, the FAI or the GDS (Tables 25–28). For the Short FES-I and the GDS, the residuals from the model showed slight violation from normality, so the models were repeated using a log-transformed outcome, but this did not change the conclusions. The results presented in the main body of the table are for the untransformed analysis, with the results from the transformed analysis included in the footer. The Short FES-I total score was categorised and is presented in Table 29. At 12 months, a similar proportion of participants in each group reported no, or low, concern about falling (intervention, 21.2%; usual care, 22.2%), but, of those who were concerned, the proportion reporting high concern about falling was slightly higher in the intervention group (32.2% vs. 30.0%). No difference in CD-RISC2 score at 6 months was observed between the groups (adjusted means: intervention 6.4, usual care 6.3; adjusted mean difference: 0.10, 95% CI –0.08 to 0.28; p = 0.26). Higher scores reflect greater resilience and adaptability.

Proportion of participants with depression

The proportion of participants who were depressed was higher in the intervention group than in the usual-care group (as measured by a score of ≥ 6 on the GDS) at all three assessment time points (12 months: 23.2% vs. 19.1%; Table 30). The adjusted OR at 12 months was 1.26 but this effect was not statistically significant (95% CI 0.91 to 1.75; p = 0.16).

Proportion of participants obtaining a fracture or multiple fractures

Over the 12-month follow-up, 31 participants (intervention, n = 17; usual care, n = 14) reported breaking or fracturing a bone as a result of a fall (adjusted OR 1.21, 95% CI 0.59 to 2.49; p = 0.60). Two participants, both in the intervention group, reported fractures from two distinct events. The types of fractures reported are presented in Table 31.

Foot pain

Participants in the intervention group reported greater foot pain at 12 months (adjusted mean 3.1 vs. 2.6; adjusted mean difference 0.43, 95% CI 0.06 to 0.80; p = 0.02). When non-compliance with the intervention was accounted for through CACE analysis, the predicted mean pain score among compliers in the intervention group was 3.1, and among the counterfactual group of compliers in the usual-care group was 2.6 (adjusted mean difference 0.50, 95% CI 0.08 to 0.92; p = 0.02).

Serious adverse events

A total of 95 SAEs were reported in the period between randomisation and 1 month following the trial end (12 months after randomisation), by 49 (9.9%) participants in the intervention group and 37 (7.2%) participants in the usual-care group (Table 32). The majority of participants (90.7%) reported only one event. During the reporting period, there were 23 reported deaths (eight in the intervention group and 15 in usual care); all deaths were considered expected. For seven deaths, the relationship to research procedures could not be assessed owing to a lack of information, but, for those that could, none was deemed to be related. Nearly two-thirds of all SAEs were hospitalisations (n = 62, 65.3%). The two events considered to be life- or limb-threatening were in the intervention group, and one of these was related to the intervention. Details of the SAEs deemed to be at least possibly related to the research are presented in Table 33. None of these events was attributable to the exercise programme but tended to relate to the trial shoes or orthosis.

Non-serious adverse events

Non-SAEs that occurred within the reporting period and were deemed to be at least possibly related to any of the research procedures are summarised in Table 34. These were all in the intervention group. Participant self-reported occurrences of pain or cramp possibly resulting from the exercises were forwarded to the treating podiatrist for review. If these events lasted for > 48 hours, then an adverse event was recorded. Pain and cramp lasting for < 48 hours was considered an expected occurrence within this population and for this component of the intervention, and so was not recorded.

Overview

Individual participant data collected in the REFORM trial were used to perform a within-trial economic analysis that comprised (1) a cost–utility analysis, in terms of the cost per QALY, and (2) a cost-effectiveness analysis (CEA), in terms of the cost per fall averted (i.e. using the primary effectiveness outcome of the trial). Costs are presented in UK pounds sterling (£) at 2015 prices, and the analysis has been undertaken in Stata version 13.1. The NICE guidelines were applied to all methods used for this economic analysis. 48

Base-case analysis

The base-case analysis was conducted on an ITT basis using multiple imputed data and from the perspective of the UK NHS and Personal Social Services, which included resource use related to falls only. The ITT aspect compares participants in the two groups (intervention vs. usual care) on the basis of their initial random allocation, irrespective of protocol deviations or withdrawal. A secondary analysis was undertaken from the societal perspective. Costs and outcome data are compared for the two groups over 12 months and, hence, discounting was not required.

Owing to the impact of missing data for the within-trial CEA, our economic analysis plan indicated that the base-case analysis would be conducted as an imputed analysis by means of MI at the utility level. This has been recommended as the appropriate method to reflect the uncertainty in the results of the economic evaluation attributable to missing data. 49

Sensitivity analysis

Additional sensitivity analyses were conducted to explore the extent to which the results change with different assumptions. Sensitivity analyses were conducted in order to test (1) complete case as an alternative method to MI for handling missing data, (2) the impact of imputing HRQoL at aggregated level (e.g. QALY level), (3) the impact of including both fall- and non-fall-related visits and hospitalisations in the calculation of total costs and (4) the societal perspective (e.g. cost of the shoes as a personal expense for the patient). Finally, we used a probabilistic sensitivity analysis to explore the uncertainty associated with the mean difference in costs and health outcomes using both the imputed and the complete data sets.

Economic data collection

Data for outcomes and resource use for the economic analysis were collected prospectively. Health service usage was measured using participant-reported questionnaires at baseline and at 6 and 12 months during the 12 months’ follow-up.

Health-related quality of life

Health-related quality of life is expressed in terms of utilities, which were assessed at baseline and at 6 months and 12 months using the EQ-5D-3L. 50 The EQ-5D is a standardised and validated generic instrument for the measurement of HRQoL that allows the translation of patient utilities into QALYs, which is the primary outcome for the base-case analysis.

The QALY is a measure of health that simultaneously incorporates changes in both morbidity (related to quality of life) and mortality (related to the quantity of years lived). As well as being one of the most widely used generic health status measures, the EQ-5D is the instrument recommended by the NICE appraisal guidance. 48 The EQ-5D considers health (functioning) in terms of five dimensions: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. Each dimension has three possible levels: no problems, moderate problems and severe problems. This five-domain and three-level system generates 245 mutually exclusive health states, including unconscious and dead. To estimate HRQoL weights (known as utilities) and to reflect the preferences of the UK population, each of these health states has been validated in a large UK population sample30 using the time trade-off method, ranging from 1 for perfect health (thus, the maximum value possible) to –0.594 for severe problems; 0 corresponds to death. Utility values were generated by valuing health status (using a social tariff) as measured using the EQ-5D system. Mean utility values were reported for each trial group, and differences in utilities between the two treatment groups were estimated using ordinary least squares (OLS) regression.

The EQ-5D has been recommended by The Prevention of Falls Network Europe Consensus as the measure of HRQoL to be used in fall prevention trials. 32 The rationale behind this is that the EQ-5D is simple and responsive to changes in health and, more importantly, it has been used widely in older populations. 51 Similarly, the EQ-5D has been used before in a UK setting to assess HRQoL and costs implications of falls in elderly people. 1 We converted the utilities derived from the EQ-5D into QALYs for each participant using the area under the curve method, following the trapezium rule, which assumes linear interpolation between follow-up points. 52 Despite the randomisation process, which should ensure that, on average, baseline variables are balanced between the groups of the trial, in practice (regardless of sample size) it is not unusual to find imbalance in mean baseline utility. As baseline utility is likely to be correlated with participants’ QALY gains over time, there are robust reasons to control for baseline utility when estimating QALYs;53 therefore, incremental mean QALYs between treatment groups were estimated with and without adjustment for baseline utility, using regression methods according to ITT allocation. In addition, incremental mean QALYs were adjusted for the same set of covariates as in the primary clinical effectiveness analysis model: age at randomisation, sex and history of falling. Centre was treated as a fixed effect in all models.

Health benefits in terms of falls

The primary outcome of the trial was the incidence rate of falls per participant during the 12 months following randomisation. The primary clinical effectiveness analysis used a mixed negative binomial regression model to analyse the number of falls per person per year controlling, as fixed effects, for sex, age at randomisation and history of falling, with centre as a random effect.

In order to interpret the cost-effectiveness results, the health outcome was reported as ‘falls averted’. The number of falls averted was estimated as the difference in mean reduction in the fall rate between the two groups in the trial estimated as per the adjusted negative binomial model used for the primary clinical effectiveness analysis.

Health-care resource use

Health-care resource use data were collected via participant-reported questionnaires. Participants were asked to complete information on their number of visits to primary care facilities (e.g. contacts with a GP and general practice nurse), use of community care (e.g. contacts with occupational therapist) and their number of hospital visits [inpatient, day case, outpatient, and accident and emergency (A&E) department] at baseline and at 6 and 12 months. Patients were also asked about the number of times they made an emergency service call or the used the Patient Transportation Service. All resource use (except inpatient hospital stay) was split into ‘fall-related’ and ‘non-fall-related’. The base-case analysis was based on fall-related resource use, except for inpatient hospital stay, given the format of the questionnaire. A sensitivity analysis explored the impact of including both fall- and non-fall-related resource use in the analysis.

Participants were asked to record the total number of times they stayed in hospital as an inpatient as well as the number of nights for each visit and the reason for attendance. The number of inpatient visits did not differentiate fall- from non-fall-related incidents and, hence, the base-case analysis included all inpatient stays reported by participants during the trial. Following a reported fall, participants were contacted to obtain information about the nature, location and cause of the fall. They were asked whether or not they sustained any injuries from the fall and, if so, whether or not they required an overnight stay in hospital. Only 28 participants (intervention, n = 14; usual care, n = 14) reported that they had to spend a night in hospital as a result of their fall. We assumed that missing answers (boxes left blank) to the second question (i.e. number of nights in hospital) when participants reported no hospital stay indicated no use of services and, thus, no overnight stays. There were participants with missing responses to number of nights but who reported the reason for their stay; for these cases the length of stay (nights in hospital) was assumed to be one night. Similarly, there were participants who reported not being in hospital but who gave information on the reason for their attendance; it was assumed that these participants stayed in hospital for one night. Finally, there were participants who reported that they stayed overnight in hospital but left blank the remaining information regarding number of occasions, number of nights and reason for stay. As a conservative assumption, we assumed that these participants stayed in hospital for one night on one occasion.

The number of visits made by the participant to their podiatry clinic in the previous 12 months was collected on the 12-month participant questionnaire. This information was available for participants in both treatment groups. In the case of participants in the intervention group, the number of visits made to the podiatry clinic as part of the intervention was collected via a trial-specific podiatrist database. Therefore, for participants in the intervention group, the number of podiatry appointments self-reported at 12 months was assumed to consist of trial appointments and unrelated routine care appointments. We knew the number of appointments received as part of the intervention and, therefore, assumed that all other reported visits were unrelated; hence, these were not included as part of the cost of the intervention.

We also asked about the use of Meals on Wheels and paid care; however, few participants reported using either Meals on Wheels (96.14% did not use this service) or paid-for help (92.87% did not pay for care). Therefore, it was decided not to incorporate these into the societal perspective analysis. Resource use was valued in monetary terms and unit costs were reported in UK pounds sterling (£) for the financial year 2014/15. The cost for each participant in the REFORM trial was calculated by multiplying health-care resource use by associated unit costs. Table 35 details the unit costs for the estimation of costs related to patient care that were used in the analysis.

Costing the intervention

The cost of the podiatry intervention was assessed based on the data collected as part of a baseline appointment questionnaire and the podiatrist database, which included information directly related to the podiatrist assessments and the intervention package received by the participant (e.g. orthosis prescription, exercise programme and exercise equipment).

Unit costs, together with their sources, for the podiatry intervention are provided in Table 36. Aside from manufacturer prices, the unit costs used in the analysis were obtained from published national sources: Unit Costs of Health and Social Care (Personal Social Services Research Unit)54 and NHS Reference Costs. 55

The base-case analysis includes only costs falling within the NHS and, hence, the cost of the shoe was not included in the primary analysis. This issue was discussed at length within the trial team and with the trial podiatrists; all were in strong agreement that, if the intervention were implemented, the NHS would not cover the cost of the shoes. We therefore considered the price of the shoes as a personal expense for the patient (e.g. as part of the societal perspective). A secondary analysis from the societal perspective that included the cost for the shoe was conducted.

We calculated the cost for each participant in the trial by multiplying their use of health-care resources by the associated unit costs. The total costs for the base-case analysis included only fall-related resource use (except for inpatient stay, which included both fall- and non-fall-related resource use). The total cost comprises five main components: (1) podiatrist visits, (2) hospital visits (inpatient, outpatient and day cases), (3) visits to primary and community health-care professionals (GP, practice nurse and occupational therapist), (4) patient transportation and (5) the cost of the podiatrist intervention. Other scenarios were tested as part of the sensitivity analysis, in which we explored the impact of incorporating both types of resource use (fall- and non-fall-related) into the analysis.

Multiple imputation

Missing data occur frequently in RCTs, irrespective of how well designed the data collection is. This is a major concern for within-trial CEA, as costs and QALYs, the main outcomes in CEAs, are cumulative measures collected over the trial follow-up. Therefore, missing data at one follow-up time point (e.g. one dimension response missing to the EQ-5D at one time point) result in missing aggregate data (e.g. total QALYs over the trial) for that participant. This problem is common in economic evaluations, as the analysis has to draw on all aspects of the study, including resource use and health outcomes. Non-response to questionnaires and returned but incomplete questionnaires reduce, often considerably, the number of data on resource use that are available for analysis. The problem is amplified when there are frequent assessments, as in the REFORM trial.

Complete-case assessment and available case analysis are proposed as useful preliminary estimations for economic evaluation but should not constitute the base case for within-trial economic evaluation. 56 An alternative method to address missing data in CEAs alongside clinical trials is MI,57 which has been recommended as the appropriate method to reflect the uncertainty in the results of the economic evaluation attributable to missing data. 49 As already stated, our economic analysis plan indicated that the base case would be conducted as an imputed analysis using MI with chained equations. Given the extent of missing data in the REFORM trial, this initial decision is justified; therefore, the base-case analysis was conducted on the imputed data set. A descriptive analysis of missing data was conducted in order to help inform the base-case assumption regarding the missing data mechanism. We described the number of missing data by treatment group at each follow-up point. We also examined the missing data pattern to find out whether or not participants with missing data were lost to follow-up throughout the duration of the trial.

Multiple imputation comprises three steps. First, the imputed data set is created through the use of regression models to predict plausible values for the missing observations from the observed values. The process includes all the variables that might be associated with the missingness mechanism (here, sex, age, history of falls, centre, baseline costs, primary care and hospital costs) and QALY utilities (at baseline and at 6 and 12 months). Costs and utilities were imputed simultaneously rather than separately in the model. Therefore, the covariates registered in the model were used for both costs and utilities, when a regression model was fitted for each variable with missing values, with the previous variables as covariates. Based on the resulting model, a new regression model is then estimated and used to impute the missing values for each variable. A random component is included to reflect the uncertainty around the predictions. Thus, MI reflects the uncertainty in the prediction of missing values while preserving the distribution and correlations in the data. 58 These values are then used to fill in the gaps in the data set. This process is repeated m times, creating m imputed data sets. It is suggested that, if missing values account for 20% of the total number of data, three imputations are sufficient. 59 Given the extent of missing data in REFORM, five imputations were performed. In the second stage, each data set is analysed independently using complete-case methods. Finally, the estimates obtained from each imputed data set are combined to generate mean estimates of costs and QALYs, variances and CIs using Rubin’s rules. 60

The correct specification of the regression model is vital to ensure that the distribution of imputed values does not differ from the observed values and, thus, that unbiased estimates are obtained. The specification of the regression models depends on the type and distribution of the variable to be imputed. Costs and QALYs are both continuous and non-normally distributed. Two alternative methods are proposed to deal with this difficulty when using MI with chained equations: (1) data transformation and (2) predictive mean matching. 61 Predictive mean matching was used for the imputation of REFORM data. This method ensures that observed data were used to estimate a predictive model (using the specified covariates) but, instead of replacing missing values with the model predicted values, the nearest observed value is used to fill the missing one. This guarantees that the imputed values are sampled from values in the original data set, and, therefore, that no imputed values will lie outside the bounds of the original data distribution. In addition to the description of the missing data mechanism, we used graphical plots to visualise whether or not the distribution of imputed data resembles the distribution of original data.

The main assumption that drives the MI mechanism is that the data are missing at random (MAR). Additional sensitivity analyses were conducted to test the robustness of the results to deviations from this assumption. In that sense we explored (1) the use of the complete data set [e.g. assume that the data are missing completely at random (MCAR)] and (2) imputation at various levels of aggregation (e.g. at utility level rather than QALY level).

Incremental analysis

Total health-care expenditure must be covered from a limited budget and, therefore, the most informative estimate for CEA is based on the difference of arithmetic mean effect from both budgetary and social perspectives. 62 Therefore, the focus of this economic evaluation was to estimate the mean costs and mean health outcomes. The cost-effectiveness of the podiatry intervention was evaluated by comparing the mean costs and outcomes (QALYs and falls) incurred in the intervention group with the mean costs and outcomes (QALYs and falls) in the usual-care group at 12 months’ follow-up, using conventional decision rules and estimating incremental cost-effectiveness ratios (ICERs) when appropriate.

As expected, costs in the REFORM trial were right-skewed, and we found that some participants had costs that far exceeded the mean value. In order to deal with skewness and heteroscedasticity, non-parametric bootstrap procedures62–64 are usually implemented as the primary statistical test for making inferences about arithmetic means for moderately sized samples of skewed cost data (such as the REFORM sample). Bootstrap methods assume that the empirical distribution of the data is an adequate representation of the true distribution of the data; the analysis is based on repeatedly sampling (with replacement) from the observed data. For the REFORM analysis we repeatedly randomly drew a sample of 1000 for each of the imputed five data sets. Each bootstrap repetition is the equivalent of a repetition of the trial. To obtain reliable results in practice it has been recommended to use at least 1000 resamples to estimate a bootstrap CI. 63 For the REFORM analysis we used 5000 resamples (bias corrected and accelerated). The mean difference in costs and QALYs for the base-case analysis was estimated using seemingly unrelated regression (SUR) equations for data on costs and QALYs. The SUR model used the same set of covariates as the mixed-effect regression model used for the clinical effectiveness analysis (sex, age at randomisation, fall history) as well as total number of falls and baseline utility. Incremental costs were also adjusted for baseline costs. SUR is used to address the correlation of standard errors between costs and QALYs. 65 This brings efficiency gains over unrelated OLS regression for three reasons: (1) it allows for explicit modelling of both costs and effects while allowing the inclusion of a set of different covariates in the two equations; (2) it exploits the existence of correlation between costs and effects; and (3) SUR does not require a new regression for every value of the cost-effectiveness threshold. 66 Again, the same set of covariates as used in the clinical effectiveness analysis was used. The baseline EQ-5D utility was also included in the utility regression to adjust for possible baseline imbalance and to reduce the standard errors of post-test EQ-5D. Incremental costs were also adjusted for baseline costs.

The ICER was estimated as the difference in mean total costs divided by the difference in mean total QALYs from baseline to 12 months. The ICER is estimated to inform decision-makers about the optimal use of NHS resources. According to standard cost-effectiveness decision rules, four different eventualities are plausible when comparing incremental costs and QALYs. If the new intervention provides better outcomes (positive incremental QALYs) at lower costs (negative incremental costs) it is considered a dominant intervention and, hence, cost-effective. If the new intervention achieves poorer outcomes (negative incremental QALYs) at higher costs (positive incremental costs) it is considered a dominated option and, hence, not cost-effective. Thus, the ICER is considered only if either intervention does not dominate, that is, both incremental costs and incremental QALYs are positive (or negative). In these last two situations, to determine whether or not the incremental health gain is worth the incremental cost, the ICER needs to be compared against a threshold value. For positive incremental costs and QALYs (the most frequent situation in HTA), an intervention will be considered cost-effective only if the ICER is lower than the threshold. According to NICE, the WTP threshold for an additional QALY ranges from £20,000 to £30,000. 48 This threshold has been used by NICE for more than a decade; however, it has recently been suggested that the threshold should be decreased to £13,000 per QALY gained. 67 According to the current established decision rules, if the result of this cost–utility analysis, namely the estimated cost per QALY, is below the £30,000 threshold, the podiatry intervention would be considered cost-effective in terms of QALYs gained.

The ICER can be rearranged in terms of net benefit, which is a more intuitive way of expressing whether or not the health benefits of the podiatry intervention are worth the additional costs. 68 The net benefit can be estimated on the cost scale as the incremental health gain, expressed in terms of money, minus the incremental cost of the intervention. The health benefits are translated into monetary value using the cost-effectiveness threshold, that is, incremental QALYs are multiplied by the WTP threshold. Therefore, the net monetary benefit (NMB) provides an estimation of the gain (or loss) in resources of investing in a particular intervention when those resources might be used elsewhere. 69 Current NICE guidance recommends presenting the NMB using values of £20,000 and £30,000 per QALY for the WTP threshold. The podiatry intervention would be considered cost-effective only if the NMB were positive.

Analysis of uncertainty

Uncertainty in economic evaluation is related to the expected values of model inputs but not to patient variability. The uncertainty around the cost-effectiveness estimates was explored by means of sensitivity analyses in order to test the robustness of the results under different scenarios. These scenarios captured variability in the estimates of costs and outcomes, which resulted from either different methods (e.g. imputation methods), from the costs included in the analysis or from the perspective for the analysis.

The extent of missing data in the REFORM trial justified the use of the imputed data set as the base-case analysis. Nevertheless, a complete-case analysis was explored as part of a sensitivity analysis. Similarly, we also conducted sensitivity analyses to test different levels of aggregation when imputing costs and QALYs.

The base case was based on the imputed data set and included only fall-related visits and hospitalisations. One-way sensitivity analyses were conducted to test the impact of including all visits and hospitalisations regardless of them having being classified as fall- or non-fall-related.

Finally, we used probabilistic sensitivity analysis to investigate the uncertainty associated with the mean difference in costs and QALYs between the two treatment groups using both the imputed and the complete-case data sets. Non-parametric bootstrapping was used to plot the joint distribution of costs and effects (QALYs) on the cost-effectiveness plane and to derive the CEAC to express the (Bayesian) probability that the podiatry intervention is cost-effective as a function of the WTP threshold. 70

Exploration of the need for a long-term model

An exploratory model was developed to explore how the differences in HRQoL observed during the trial (e.g. at 1 year) evolve beyond the study (up to 5 years). For this exploratory projection, we used a decision-modelling approach and assumed that the difference in HRQoL and costs observed at 1 year would remain unchanged.

Validation of results

In order to validate the results of the analysis, two statistical codes (written in Stata) were independently developed and their results compared. The codes were developed by one analyst and checked independently by another. The distributions of the observed and imputed values were compared graphically.

Patient population and missing data mechanism

Twenty-four participants died during the trial: 9 out of 493 (1.8%) in the intervention group and 15 out of 517 (2.9%) in the usual-care group. When there were missing data before these participants’ deaths, the imputation process was applied in the same way as for the rest of the patients in the trial. The questionnaires that should have been received at any other assessment after their deaths were considered as part of the complete-case analysis with zero resource use and zero utilities. The complete-case analysis comprised those participants for whom data were available for the whole trial duration for utilities and all cost categories.

The proportion of participants with complete data decreased with the duration of follow-up but remained similar in both groups (Table 37): from 72.0% (baseline) to 54.4% (12 months) for the intervention group and from 71.8% (baseline) to 61.3% (12 months) for the usual-care group. In the usual-care group, more individuals are observed at 12 months than at 6 months; therefore, the missing data follow not a monotonic pattern but an intermittent one (i.e. there are participants with missing 6-month data but complete data at 12 months). A complete-case assessment would be, as a minimum, inefficient because it would discard observed data from individuals with some missing outcomes.

Table 38 presents the ORs from a logistic regression of indicators of missing QALY data on treatment group allocation and the covariates used for the main statistical model. Lower EQ-5D at baseline is associated with missing QALY data but is not statistically significant at the 5% level. This suggests that the data are unlikely to be MCAR. It was found that older and frailer (lower utility) participants with a history of falling were more likely to have missing QALY data. This information would support the MAR assumption (e.g. MI assumed missing data mechanism).

Multiple imputation and likelihood-based methods can handle non-monotonic missing data under the MAR assumption while incorporating the uncertainty around the unobserved data and maintaining the correlation structure. 56 Therefore, the base case for the REFORM analysis uses MI. A complete-case analysis, which is not valid under MAR, is presented only for comparison.

The MI model was validated by comparing the distributions of the observed and the imputed data (Figures 6 and 7). The distributions of imputed data are similar to the distribution of the observed data. The MI data sets were analysed with the same SUR model used for the complete-case analysis.

FIGURE 6.

Comparison of the distribution of imputed values (imputations 1–5) with the observed data (imputation 0) for total costs.

FIGURE 7.

Comparison of the distribution of imputed values (imputations 1–5) with the observed data (imputation 0) for total QALYs.

Health-related quality of life

The complete-case analysis for utilities consisted of the participants who returned all questionnaires and completed the EQ-5D questions. The EQ-5D is classified as complete only if its five dimensions contain a response. Table 39 shows the number of questionnaires returned (including those with missing dimensions) and the number of completed EQ-5D questions for each time point. The number of questionnaires returned decreases with time.

Table 40 describes the number and proportion of participants in the REFORM trial reporting each of the levels on each EQ-5D dimension.

At baseline, participants reported problems in mobility and pain more than in the other dimensions. These domains are worse for the participants in the intervention group: mobility (56.9% usual care vs. 59.7% intervention) and pain (56.6% usual care vs. 78.4% intervention). As expected, the intervention improved mobility as data showed an 11% reduction in the number of participants reporting problems from baseline to 12 months in the intervention group (compared with 1% change in the usual-care group). Similarly, the intervention improved pain (15% reduction in participants reporting problems in the intervention group compared with 10% change in the usual-care group).

The likelihood of remaining in perfect health decreased with time (Tables 41 and 42). However, the reduction in the number of participants in perfect health is lower in the intervention group (7.4%) than in the usual-care group (17.7%). The data also suggested that improvement in anxiety/depression is proportionally even greater than the improvement in other dimensions, especially among participants in the intervention group (19% reduction in numbers reporting anxiety problems).

Utility values were generated by valuing health status (using a social tariff) as measured using the EQ-5D system. The analysis of utilities (Table 43) shows that participants in the intervention group start from a lower baseline utility, on average (0.67 for the intervention group vs. 0.69 for usual care). The data also showed that participants in the intervention group had, on average, 0.14 [standard deviation (SD) 0.38] admissions to the hospital during the 6 months before randomisation, whereas participants in the usual-care group had, on average, 0.12 (SD 0.45) admissions. This emphasises the need to adjust the utilities gained for baseline utility level. When considering this baseline imbalance, the data show that by the end of the trial participants allocated to intervention obtained, on average, a marginally higher HRQoL gain than participants allocated to usual care.

The overall distribution of EQ-5D scores (utilities) for the different follow-up time points is illustrated by treatment group in Figure 8. Utilities at baseline ranged from –0.181 to 1 for both groups; at the end of the trial, utilities ranged from –0.239 to 1 for both groups.

FIGURE 8.

Distribution of EQ-5D scores by treatment allocation and time point. (a) Baseline: usual care; (b) baseline: intervention; (c) 6 months: usual care; (d) 6 months: intervention; (e) 12 months: usual care; and (f) 12 months: intervention.

The distribution of mean utilities across the 12-month follow-up for the two groups is shown in Figure 9. The usual care participants reported higher HRQoL at baseline and at 6 and 12 months. All of the differences were small, and the 95% CIs overlap at each time point.

FIGURE 9.

Mean EQ-5D scores at baseline and follow-up time points by treatment group.

The mean QALYs were estimated based on individual participants’ utilities. Table 44 summarises the mean QALYs and the difference between treatment groups for all available cases. At the end of the trial, participants allocated to the intervention obtained, on average, a marginally higher QALY gain than participants allocated to usual care (Figure 10) when adjusted for baseline utility (0.010 QALY gain). The difference is 0.083 when adjusted for all covariates.

FIGURE 10.

Mean total QALYs for the intervention and usual-care groups.

Health-care resource use and costs

The mean levels of resource use over the trial based on all available data are shown for the two treatment groups in Table 45. Although participants in the intervention group had, on average, fewer hospital day cases and used the patient transportation service fewer times, they had, on average, more hospital admissions, more outpatient visits and more A&E attendances than usual care participants over the trial duration.

Participants in the intervention group had, on average, fewer visits to the GP than the usual-care group (Table 46); however, they had, on average, more visits to the practice nurse and the occupational therapist. Participants in the intervention group undertook more podiatrist visits in total than the usual-care group, as this group received this service as part of the intervention.

Table 47 summarises the mean cost by item of resource use based on all available cases and according to treatment group. Costs associated with hospital inpatient stay and the intervention itself were the major cost drivers for the participants in the intervention group.

Twenty-eight participants (intervention, n = 14; usual care, n = 14) reported that they had to spend a night in hospital as a result of a fall. The average cost per inpatient stay was £7121.00 (SD £1535.83) in the intervention group and £6666.50 (SD £1156.84) in the usual-care group. Therefore, inpatient stay based on fall calendars was £454.50 more expensive for participants in the intervention group (95% CI –£601.80 to £1510.80). Given the small sample size, it was decided to estimate inpatient stay based on data from the participant questionnaires. The only limitation of this is that the participant questionnaire did not differentiate inpatient stay as fall- and non-fall-related, although the rest of hospital stay (outpatient, day case and A&E) did differentiate between these.

Cost–utility analysis and uncertainty

The base-case analysis (Table 48) shows that the participants who were randomised to the intervention experienced (marginally) improved health outcomes. At the end of the trial, the intervention group had experienced 0.0129 (95% CI –0.0050 to 0.0314) more QALYs. However, the intervention is more costly than usual care [on average £252.17 more per participant than usual care (95% CI –£69.48 to £589.38)] when adjusted for all covariates (including baseline utility). ICERs ranged between £19,494 and £20,593 (societal perspective adjusted for all covariates) per additional QALY. For both the base-case and secondary analysis (societal perspective), the probability of the intervention being the more cost-effective option is > 0.60 for the incremental analysis adjusted for baseline EQ-5D, and > 0.65 when incremental QALYs are adjusted for all covariates. The NMB associated with the intervention is positive, indicating that the intervention is cost-effective, as the resources to be displaced would be less than the benefit to be gained if the intervention was implemented in the NHS. However, these results were calculated from the point estimate of the difference in QALYs; the lower-bound confidence limit for the 95% CI was negative, and, therefore, there is the potential for a negative QALY gain.

TABLE 48 - Summary for incremental analysis (ITT), cost-effectiveness results and uncertainty for the base-case analysis (highlighted) and sensitivity analyses

Difference between groups (intervention – usual care) and 95% CIs were estimated from a bivariate model using SUR. The covariates used to adjust for in the model were age, sex, treatment group, baseline utility and history of falling.

Analysis	Difference in costsa (95% CI)	Difference in QALYsa (95% CI)	ICER for the intervention (£ per QALY)	Probability intervention was cost-effective £30,000/QALY (%)
Base case (MI), NHS perspective	252.17 (–69.48 to 589.38)	0.0129 (–0.00 to 0.03)	19,494.35	65.58
Sensitivity 1 (complete case)	272.86 (–349.6 to 916.56)	–0.0091 (–0.04 to 0.02)	Intervention dominated	17
Sensitivity 2	222.34 (–156.6 to 605.1)	0.0109 (–0.007 to 0.029)	20,385.75	61
Sensitivity 3	441.88 (–273.1 to 1052.4)	0.0150 (–0.002 to 0.033)	29,454.34	49
Sensitivity 4	327.17 (–65.17 to 451.09)	0.0140 (–0.003 to 0.032)	23,341.72	60

The incremental cost-effectiveness plane (Figure 11) demonstrates the uncertainty associated with the mean difference in costs and QALYs between both intervention groups by plotting the non-parametric bootstrapping results. A total of 5000 bootstrapped replicates of differences in costs and QALYs are shown. The majority of the replicates falls within the north-east quadrant, indicating that the intervention is more effective but more costly.

FIGURE 11.

Cost-effectiveness plane (ITT analysis) for the base-case analysis: base-case NHS perspective (MI adjusted for baseline utility).

The CEAC derived from the joint distribution of costs and effects is represented in Figure 12. The curve was constructed by plotting the proportion of incremental cost–effect pairs that are cost-effective for a range of thresholds. The horizontal interrupted line indicates a 50% probability of the intervention representing value for money for the NHS. The probability of the intervention being cost-effective is > 60% given the current NICE WTP threshold of £30,000 per additional QALY.

FIGURE 12.

Cost-effectiveness acceptability curve (ITT analysis) for the base-case analysis: base-case NHS perspective (MI adjusted for baseline utility).

Cost-effectiveness analysis and uncertainty

A cost-effectiveness analysis, in which the outcome is expressed in terms of number of falls averted, may be more intuitive to interpret for health-care professionals. However, there is no established WTP threshold for an additional fall averted. Therefore, the cost per fall averted was assessed for comparison, as allocation decisions can be made based only on cost per QALY estimates. In the base case, the podiatry intervention was both more costly (mean incremental cost £241.64, 95% CI £–98.08 to £581.37) and more effective (mean incremental effect 0.19 falls averted per person year, 95% CI –0.05 to 0.44 falls averted per person year), with an incremental cost per fall averted of £1253.82 (ICER); however, for both of these parameters the lower 95% confidence limit is negative, and so this does not exclude the possibility of a negative result. Figure 13 shows the incremental costs and incremental effects on the form of a cost-effectiveness plane. There is significant uncertainty in the effectiveness estimates, as the estimates fall on both sides of the x-axis. Figure 14 shows the CEAC per fall averted.

FIGURE 13.

Cost-effectiveness plane (ITT analysis) per fall averted.

FIGURE 14.

Cost-effectiveness acceptability curve (ITT analysis) per fall averted.

Sensitivity analysis

Handling missing data

The complete-case analysis was tested as an alternative method to MI for handling missing data. The complete-case scenario comprised 286 (28.3%) participants, of whom 129 (26.2%) were in the intervention group and 157 (30.4%) were in the usual-care group. The complete-case analysis shows that the intervention group accumulated greater costs and reported lower HRQoL than participants randomised to usual care. The intervention costs were, on average, £272.86 more per participant than usual care (95% CI –£349.63 to £916.56), although accumulated total QALYs are smaller than those for usual care (mean difference –0.0091, 95% CI –0.0396 to 0.0196). Therefore, complete-case results indicate that the intervention is dominated by usual care (Figure 15). The NMB associated with the intervention indicates that the resources to be displaced would be greater than the benefit to be gained if the intervention were implemented in the NHS. Figure 16 represents the CEACs for the complete case when adjusted for all covariates. The probability of the intervention being cost-effective is < 20% given the WTP for an additional QALY up to £30,000; therefore, the intervention is unlikely to be cost-effective based on a complete-case analysis.

FIGURE 15.

Cost-effectiveness plane (ITT analysis) for the complete-case analysis (NHS perspective, adjusted for covariates).

FIGURE 16.

Cost-effectiveness acceptability curve (ITT analysis) for the complete-case analysis (NHS perspective, adjusted for covariates).

Given the accumulative nature of costs and QALYs, these variables can be dealt with at different levels of aggregation. The base-case analysis estimated QALYs by imputing missing utilities (e.g. disaggregated level). A sensitivity analysis was conducted for the base case on the imputed data set in which we explored the impact of imputing HRQoL at QALY level (e.g. aggregated level). Imputing HRQoL at an aggregated level has no impact on the cost-effectiveness of the intervention).

Assessing and ordering footwear during usual practice

Podiatrists reported how patients frequently wore inappropriate footwear and cited common issues seen during practice as narrow shoes, inappropriate heel height or shoe style and a lack of appropriate fastening. Although some issues with men’s footwear were reported (e.g. fastening and inappropriate slippers), it was the perception of the podiatrists that the majority of issues were with women’s footwear:

It tended to be the women that weren’t wearing the sort of shoes that would balance and that were comfortable, they tended to wear, I don’t know, more fashionable shoes forgetting about what sort of age they were and they hadn’t thought about the fact that the shoes were possibly causing their instability.

Podiatrist 7

All podiatrists discussed how general footwear advice was provided as part of their usual practice, with a small number of podiatrists basing this advice on falls prevention. Routinely, footwear advice involved discussions around the types and styles of shoes and placed a particular emphasis on the importance of indoor shoes. Central to these discussions was the need to enhance the patient’s understanding of ‘good footwear’. To facilitate this, podiatrists described how they often directed patients to cheaper footwear alternatives or specific companies, to demonstrate that a good shoe was not necessarily reflected by its price. In addition, podiatrists spoke of using prompts such as shoe catalogues, leaflets and sample shoes that were compared with patients’ current footwear, in order to further patients’ understanding.

Variation in the criteria to assess patients’ footwear was reported within routine practice; for instance, some podiatrists described how they made assessments by glancing at the patients’ footwear or how the assessment was ‘second nature’. The characteristics of footwear commonly assessed during routine practice tended to include fastening, length, width, heel height, sole and fabric. Podiatrists confirmed that there is currently no formal checklist used within routine practice for footwear assessment:

Yes it’s in your head and then you tick a box on SystmOne to say appropriate footwear worn . . . it’s second nature.

Podiatrist 5

Podiatrists reported that footwear was not routinely provided, with only one podiatrist (podiatrist 5) referring to fitting ‘stock hospital shoes’ when there was a clinical need.

Experiences of assessing and ordering footwear during the REFORM trial

For the purposes of the REFORM trial, participants were asked to bring samples of their indoor and outdoor footwear to the clinic for assessment. As described in Chapter 2, podiatrists were asked to assess participants’ footwear against a checklist provided by the research team of characteristics of suitable footwear identified in the literature12 (see Appendix 21). If trial participants failed to bring samples of their footwear, the assessment was based on a description. The majority of the podiatrists found the trial criteria for assessing footwear straightforward to follow and described the checklist as clear and logical:

It [the checklist] was very similar to what I would normally do and it actually reiterated all the things that I was doing before so it was, you know, clarified everything that I was doing and thinking yep that it was [what] I would do anyway, so I mean that was quite useful because it took me back to basics to actually think about it a bit more.

Podiatrist 7

However, a few podiatrists mentioned that the current electronic medical record did not have sufficient data fields for the more detailed footwear checklist; therefore, alternative ways of capturing these data would have to be found:

In our paperwork we have whether it’s a good fitting shoe, whether it was too big or too small, whether it’s a slipper or bespoke orthopaedic aid and then we have an optional heel height and what kind of fastening they have and we do have a comment box as well. So that’s our kind of footwear assessment of what we can document.

Podiatrist 10

The REFORM trial also enabled podiatrists to order footwear for participants, which is something that is not currently provided in usual podiatry care. Prior to ordering shoes, podiatrists were required to measure participants’ feet. The measuring guide was a laminated picture of a foot, annotated with different shoe sizes. Participants stood on the guide and their corresponding shoe size was read. The width of shoe was determined by measuring the circumference of the foot using a tape measure and referring to a chart to find the required width (www.dbshoes.co.uk/measuring_chart.php, accessed 2 October 2012; www.hotter.com/gb/en/info/Hotter-Shoes-Fitting-Guide, accessed 2 October 2012). Although the majority of podiatrists described the measuring process as straightforward, a number of difficulties were reported, especially among those who did not have training in shoe fitting outside the trial (the trial did not provide additional training on this). Some podiatrists commented that the measuring guide on which participants stood during the measuring process would sometimes slip unless it was taped to the floor. Challenges for less mobile patients were also reported, as they found it hard to stand on the measure, which had to be placed against a wall.

Despite some difficulties, the majority of podiatrists felt that footwear assessment and advice was generally straightforward and a central element of their clinical role; however, some expressed concerns that in routine practice there would be insufficient time to undertake a full footwear assessment:

It is time-consuming but we were given enough time to do it, so in an ordinary clinic if you had 20 minutes to do a routine treatment and educate them on footwear and have a discussion around it, it would really eat into the time but for the study we had enough time to do that.

Podiatrist 1

Given these time pressures within usual practice, a number of the podiatrists did suggest that this element of the intervention could potentially be conducted by a podiatry assistant or technician:

. . . probably our technicians or podiatry assistants are probably very, very competent at doing that type of thing.

Podiatrist 7

In contrast, however, one podiatrist (podiatrist 4) felt very strongly that footwear measurement was outside the role of podiatrist and should be conducted by an orthotist, working as part of a multidisciplinary team.

Additional issues with the measuring process included the accuracy of the sizing guide, fitting slippers that were available only in full sizes (not half sizes), the lack of footwear under size two and the time required to fit footwear:

That was out of frustration because of the backwards and forwards process of ordering, fitting and finding out that the shoe wasn’t right.

Podiatrist 3

A minority of podiatrists also found the shoe ordering process very time-consuming, something that was attributed mainly to patients spending large amounts of time selecting footwear from what some podiatrists considered an excessive number of options.

Experiences of and adherence to footwear advice/trial purchased shoes

As would be expected, adherence to footwear advice varied among participants. From the 12-month follow-up questionnaire administered to the whole of the trial intervention group, we observed that nearly two-thirds (n = 137, 63.7%) of the participants who reported that they had their shoes checked said that their podiatrist gave them advice about their footwear or suggested that they should wear a different style of shoe, of whom 104 (77.0%) reported that they followed this advice (80.8% of men and 74.7% of women).

It was the view of both podiatrists and trial participants that it can be difficult to action footwear advice and change shoes/slippers, as high-street shops do not always stock footwear that is a suitable fit, and in some areas there is a lack of stockists:

But you see they say what makes an everyday shoe unsafe . . . what they mean about secure fastenings but they’re not easy to find either . . . That’s what I need, depth for my toes and they have depth but you don’t get a great deal of choice and I believe there’s a shop down [name of street where shop is] that deals in, just a little shop and they deal in shoes, I haven’t been in. I’ve just looked in the window as I’ve passed by.

Trial participant 8

Several podiatrists highlighted that, for many service users, the cost of the appropriate footwear recommended could be prohibitive:

Obviously these brands that you recommended, you know, in the shops they can be 70 to 80 pounds, so you know, unfortunately some people just can’t afford to get more specialised footwear.

Podiatrist 6

The availability of shoes through the trial had provided respondents with access to appropriate footwear by overcoming these barriers. In addition, podiatrists were at pains to discuss with service users that, although many shoes were expensive, ultimately the cost of the shoe did not always reflect whether or not it was appropriate for them:

. . . so I think [the service providing shoes] is a great idea in theory but I think it’s probably quite costly and maybe, you know, you can buy, you know, if you’ve got to buy a pair of shoes there’s a lot of shoes you can buy that aren’t as expensive but still have those good qualities, you know, like a cheaper pair of trainers really.

Podiatrist 12

The response to the footwear advice/footwear provided was mixed. Many participants reported that they wore their ‘appropriate’ outdoor shoes all the time and were really satisfied with the fit and choice of shoe:

I think I’ll always wear Hotter shoes now because they’re so good.

Trial participant 2

Other participants described how they stopped wearing the footwear, largely because the shoes did not fit properly and were uncomfortable; in some cases, even the recommended shoe suppliers did not have shoes that were a good comfortable fit for people who are older (some of whom had various foot problems including arthritis, bunions or corns). In addition, there were also pragmatic reasons why appropriate shoes were more difficult to identify for some service users, such as difficulty in putting shoes on among this age group:

The age group of the patients in the REFORM trial were very elderly, or a lot of them were, and for that reason they had difficulty getting down to their feet, so they tend to go for slip-on shoes for ease of fitting them whereas a lace-up or Velcro-fastening shoe created additional problems.

Podiatrist 3

Despite participants demonstrating a good general understanding of ‘appropriate footwear’, there were still a number of examples of inappropriate footwear being worn, reflecting the pragmatic solutions the participants had found in order to incorporate the advice into their daily routines and practices:

I think if you wear sensible shoes, I think that’s the big thing. I mean I’m past all these stiletto heels now. I used to wear them once but no, I do wear a little heel sometimes when I go out. I like these flip-flops in the summer because they’ve got the heighted heel at the back and I find that very comfortable. I’m not very keen on dead flat but because of my neck I don’t like a very flat shoe. Yeah I think sensible shoes.

Trial participant 6

Although not particularly supported by the quantitative data, podiatrists perceived that women were less likely to action the footwear advice and that this needed to be taken into account when such advice was being provided:

It’s about the complete loss of identity . . . you have to try and see if from the patient’s point of view because it’s very cut and dry for us. This is what you need, this is what’s going to be really good for you and that’s what you’re going to get and sometimes you have to work a little bit more around the patient and I think, you know, going back to the original discussion about footwear, I think women have an idea about how they want to present themselves to the world and if say that foot support or orthotic is not going to fit in that shoe or the shoe is not going to support the function of the device, it can be very difficult.

Podiatrist 4

Using orthoses during routine practice

The majority of podiatrists described how they had prescribed some form of orthosis in their usual practice, predominantly for clinical conditions such as Achilles tendon problems or plantar fasciitis; however, one podiatrist (podiatrist 1) reported that they had prescribed an orthosis for a participant who had fallen. Of the podiatrists who had prescribed orthoses, the majority had prescribed the trial orthosis (X-Line) or a similar orthosis. Issues encountered when routinely prescribing orthoses were reported and included complaints that they were uncomfortable or too bulky to fit in participants’ shoes:

Sometimes they [the participants]find them a bit too bulky, so they can’t get their foot in as well as the insole and just to be able to get them in the shoes themselves.

Podiatrist 2

Experiences of the X-Line orthosis during the REFORM trial

The majority of podiatrists and trial participants reported positive experiences of using the X-Line orthosis and described the X-Line as a good, cost-effective orthosis. Podiatrists also explained how the X-Line was good for participants who had arthritis or deformities, as the orthosis was slim and so could easily fit into footwear. The arch support, control on the heel and met dome support were also cited along with the fact that the X-Line was not overly corrective and so was not expected to cause problems such as lesions or balance problems. Direct comparisons with other routinely prescribed orthoses were also made, with podiatrists often stating a preference for the trial orthosis. Podiatrists’ positive experiences of the X-Line orthosis were also demonstrated when one site reported that it may consider changing its routinely prescribed orthosis to the X-Line, although this would ultimately depend on the cost of the device and whether or not those who purchased equipment in the trust would agree to order this type of device:

These weren’t bad at all because they’re quite thin on the front as well, that’s another problem we usually have of patients especially if they’ve got arthritis in their toes, claw toes that type of thing but even the slight raise sort of lifts the foot up and then the shoes are pressing on the toes. But no I was quite impressed with those.

Podiatrist 8

On a practical level, the majority of podiatrists found the X-Line insoles easy to fit into participants’ footwear. Trimming the insole was generally not found to be an issue, as this could mainly be done at the clinic with scissors. If required, the insole could also be trimmed to three-quarter length, to aid fitting into participants’ footwear. Attaching postings and additions was also not perceived to be a problem, as they were self-adhesive. This meant that they could be attached in the clinic and so did not require special equipment or to be sent to the laboratory, as is sometimes the case:

The additions we just had the peel back, so like the sticky on the back and you just peel it off and away you go whereas the ones we use at the minute, they have to go to the lab and like be glued on. So obviously this is easy because you’ve got them in clinic and you can put them on there and then and off they go.

Podiatrist 2

However, although the majority of podiatrists related positive experiences, theere were some difficulties reported with fitting and trimming the X-Line. For instance, one podiatrist (podiatrist 3) stated that trying to fit the orthosis into a participant’s footwear made the shoes too tight and inappropriate:

The only difficulty came in with trying to fit an orthotic [insole] to what was an appropriate shoe and if the shoe then became too tight then it was no longer an appropriate shoe.

Podiatrist 3

At study set-up, some sites raised concerns about the possibility of the trial identifying patients with an unmet clinical need, who would require a full biomechanics assessment. Sites were apprehensive about the impact that this would have on their clinics. These concerns were not represented within the interviews; the majority of podiatrists were willing to prescribe the insole without a full assessment, as they considered the device unlikely to cause participants any problems. However, three podiatrists reported giving participants an assessment to check that it was clinically appropriate to prescribe the orthosis:

Yeah I mean it was a slight assessment with checking muscle strength and basic things but I wouldn’t say it was a full complex biomechanical assessment.

Podiatrist 8

Experiences of and adherence to the orthosis prescribed during the REFORM trial

For those in the trial as a whole, Table 51 presents responses to the adherence questions for the intervention participants who received an orthotic insole and responded to this question. At 12 months, 66.4% of participants reported wearing their orthosis most or all of the time, and 85.0% reported wearing it at least a little of the time.

It is clear that wearing orthoses was generally acceptable; however, the podiatrists adapted their use to the individual circumstances of the participant, for example by adapting the insole to accommodate foot deformities. As might be expected, how comfortable the participant found the insole was a significant determinant in whether or not it was worn with any regularity:

They’re comfortable and that’s the main thing as well, if they weren’t comfortable they would have been chucked out.

Trial participant 7

However, those who felt that the orthosis was likely to have a benefit, through either previous experiences of using orthoses or anecdotes from family or friends, seemed to be prepared to endure some discomfort, at least initially. Some participants were also willing to persevere with the orthotic, despite some initial discomfort, because it had been recommended by the podiatrist and they felt that it was likely to be of benefit. Several reported that they considered the insoles to be a good idea, especially as they had either fallen several times or because they wanted to find something to help improve their walking, and demonstrated a willingness to try:

Well it took several weeks to get accustomed to the insoles, it was quite painful but I was determined to persevere.

Trial participant 4

Although in some cases the perseverance paid off, others ceased to wear the orthosis completely:

I used the orthotics, I had some from before, and they are not comfortable with my back. I mean there’s nothing much the matter with my back but I ended up with a sore back, with wearing them.

Trial participant 13

A contributing factor in the resultant comfort was whether or not appropriate footwear was available. Although the trial had attempted to accommodate the requirement for footwear when necessary, for some participants, even with this option, the orthosis was still not a comfortable fit. This was especially the case for individuals who had pre-existing problems with their feet, which made it difficult to cope with the inserts, or who had to adapt the use of the inserts to make them tolerable:

Yes but on one foot I could wear them but not on the other because I’ve got very high arch, I broke this foot when I was very, very young and it just made a difference. It made the shoe too tight.

Trial participant 3

However, for many the whole package of obtaining the appropriate footwear and orthosis worked well and participants could feel the benefit from the increased support. Positive experiences were mainly associated with comfort and support, for example improvement in posture or shoes fitting better and being more comfortable to wear. This was in addition to the perceived effect it had on participants’ balance, the number of falls they had and their confidence:

. . . [the podiatrist] decided the shoes, although they were Clarks [C&J Clark Ltd, Somerset, UK] the shoes I was wearing, they weren’t as good so she measured me for some shoe inserts and I got a pair of Hotter shoes and a pair of Hotter slippers and these shoes and the inserts they’ve made a dramatic difference, you know, to me walking and lifting; because I think my arches have probably fallen a bit, so it gives me support in that way.

Trial participant 2

Those who were experiencing a noticeable benefit tended to wear the orthoses all the time:

I’ve got one set for these shoes and then another set I use if I go out anywhere or leisure, I just slip them into whatever shoes I’m using.

Trial participant 2

However, others, who did not feel such a palpable benefit or who did not have the expectation that orthoses were likely to affect their risk of falls, tended to use the inserts in a more pragmatic way, and adapted their use to what was practical and feasible:

I’ve got them in one of the pairs of shoes that I wear most of the time . . . [I wear them] probably four times a week, because I can’t get them in, like I say, I’ve got a problem with my toe, so if I can’t put them into any of my other shoes because there’s not enough room in there for my toes and that in sole . . . I can’t say that I’ve really felt a difference, do you know what I mean – I’ve got used to wearing them so they’re very comfortable.

Trial participant 7

Prescribing exercises in routine practice

Some podiatrists reported prescribing exercises in their routine practice, largely for conditions such as plantar fasciitis or Achilles tendon injuries. The podiatrists were not currently prescribing exercises for falls prevention in routine care. They acknowledged issues associated with prescribing exercises, which included the potential for the exercises to cause injury:

Yeah but with older people a lot of the exercises you’ve got to be really careful with that you don’t cause further problems. Some of the exercises I think the patients go a bit too far with them and would actually sort of damage tendons if they overstretch.

Podiatrist 8

Exercises prescribed during routine practice differed from those in the REFORM trial. Although podiatrists reported having prescribed some, if not all, of the exercises provided during the trial, these exercises had not been prescribed as a ‘package’ or in combination with each other. In terms of the individual exercises, resistive bands and foot therapy balls were rarely provided but may have been prescribed in biomechanics clinics. Some podiatrists raised the issue of professional domains, remarking that, in current practice, prescribing exercises is the responsibility of physiotherapists rather than podiatrists:

. . . but we’ve never had therabands within podiatry stock.

I think some of it is expensive, is that the reason?

I think that would be, yeah the main reason I would think and it just seems to be for our trust, it seems to be the role of the physiotherapist. So we maybe refer patients to physiotherapy for that part of an exercise programme but we haven’t done it within the podiatry clinics.

To negate the need for specialist equipment in routine practice, podiatrists would suggest alternative ways of conducting exercises; for example, they might advise patients to use a dressing gown cord instead of a resistive band. In addition, over half of the podiatrists reported that they had prescribed exercises using alternatives to a therapy ball that required patients to roll their feet over a can, or to pick up pencils or golf or tennis balls.

Experiences of the exercises during the REFORM trial

Podiatrists mostly spoke positively of the trial exercises. Indeed, two podiatrists reported having changed their routine practice to prescribe exercises when appropriate. However, some practical issues were reported, including the difficulty with prescribing exercises for elderly and frail patients, especially given the number of exercises that were included and the length of time it took to explain them. Podiatrists took a pragmatic approach and modified the trial exercise package to adapt the regime for people with comorbidities, to reduce the number of exercises or to suggest that particular exercises were not undertaken:

Yes when we were finding exercises for the very elderly patients, a lot of them struggled to actually be able to go on tiptoe, stand on one foot because they were quite frail. Some patients also had other health problems which made exercise a lot more difficult such as osteoarthritis of the feet, so they would often phone up and say, I’ve tried very hard to do these exercises but they’re causing me a lot of pain.

Right and so in those cases what was the advice you gave them?

We asked them to modify their approach and to do as many of the exercises as they could do but not feel too bad if they had to reduce the frequency of the exercise or maybe even miss one exercise out, for instance, if they had osteoarthritis of the first MTP [metatarsophalangeal], big toe joint, we would say to them, don’t worry about getting on tiptoe but do and try and do the other ones.

To facilitate trial participants’ understanding of and adherence to the exercises, a booklet and a DVD were provided. Podiatrists and trial participants spoke positively of the booklet, describing it as well written and easy to follow, with clear instructions and a good combination of pictures and text. The booklet was viewed as a useful resource that helped podiatrists to remind and explain to patients how to do the exercises, something that was considered particularly important given the amount of information patients received at their first appointment. During the trial, podiatrists used the booklet as a guide when delivering the whole intervention and also worked through the booklet with patients, using the pictures to aid their description of the exercises during clinic. Although one podiatrist (podiatrist 3) felt that some participants viewed the information in the booklet as too simplistic, others saw its simplicity as a strength, stating that it could have been issued without additional instructions. The positive feedback provided by podiatrists is exemplified by the fact that a few sites asked if they could give out the booklet in their routine practice prior to the end of the trial, with one podiatrist (podiatrist 10) also presenting the booklet at a staff meeting:

I thought it was very good. It was very well written. It was very easy to follow and because it had the pictures, the people could, you know, go back to it and have a look and it did make sense. It was written well.

Podiatrist 7

In light of podiatrists’ initial concerns regarding the amount of information given to trial participants at the start of the study, both the DVD and the booklet were considered helpful reminders. However, only a small proportion of trial participants reported having watched the DVD. The podiatrists provided some insight into this during their interviews, as they reported that patients had difficulty playing the DVD or, in some cases, did not have a DVD player. In addition, one podiatrist (podiatrist 3) suggested that the DVD may be more suitable for younger people and felt that those in the age group taking part in the trial would prefer written information:

Some of the DVDs did not work and not every patient had access to a DVD player and the patients tended to report back that while they found the booklet very helpful because that generation tends to like to read things rather than play things, if you’re dealing with a younger generation the DVD would have been more useful.

Podiatrist 3

Podiatrists felt that a follow-up appointment was important to ensure that the exercises were being conducted correctly:

. . . showing the patient how to do them correctly and getting them to do them correctly, because occasionally when they came back to the clinic for their review appointment and you asked them to demonstrate the exercises again, some of them hadn’t been doing them correctly, so it just needed a bit of re-education really.

Podiatrist 1

Experiences of and adherence to the exercises prescribed during the REFORM trial

At 12 months, 28.9% of participants in the intervention group reported performing the exercises at least three times per week and 74.5% reported performing them at least once per week (Table 52).

In the interviews, participants reported varying levels of adherence to the exercises, with some reporting undertaking them every day, some reporting undertaking them three times a week and some reporting that they did not do them at all. The length of time that participants complied for also varied, with some reporting trying the exercises for a short time before stopping (e.g. 1 month), although others persevered because of their desire to prevent falls.

In the main, the trial participants spoke positively about the trial exercises, although, in a similar vein to the podiatrist interviews, the number and challenging nature of some exercises was discussed. Trial participants and podiatrists expressed varying opinions regarding the resistive band and therapy ball exercises, which are outlined in Table 53.

Although some participants reported that they had not noticed any differences since undertaking the exercises, for many participants the exercises had led to perceived benefits, such as improvements in their walking, balance, confidence and body awareness:

I didn’t do them every day to start with but I did them at least three times a week and within several weeks I felt the benefit. I did feel the benefit from, you know. In my walking and in my balance and the ability to get to stand up.

Trial participant 2

Whether or not participants perceived the exercises to be beneficial may have influenced their adherence. For example, one participant commented that they had continued to undertake the exercises even when they were tired, as their perception of improvements had given them an incentive to continue.

A number of participants also described strategies that they had adopted to make the exercises easier to fit into their daily lives, for example goal-setting (e.g. wanting to stand up easily), splitting the exercises throughout the day, and doing the exercises ‘first thing in the morning’ or while watching television, which may have improved adherence:

I do them three times a week all at once, one after the other. Monday, Wednesday and Friday and I’ve done them this morning before 7 o’clock. Well I make sure because I do it at a time that suits me that doesn’t interfere with my life, you know, because I’m out of the house by quarter to nine every morning, so you know, that’s the time it’s done.

Trial participant 1

For participants who did not regularly undertake the exercises, a range of reasons for non-adherence were provided, including the length of time required to undertake them. Medical conditions such as heart problems or arthritis also made performing the exercises difficult or painful for a number of participants:

. . . because I’ve got a bad heart, it doesn’t take me long to get out of breath, so I found them hard work, very hard work.

Trial participant 7

Others had to adapt the exercise package to accommodate their own physical limitations:

One I couldn’t do. The one where you had to stand up with your back up against a wall. I couldn’t do that because I couldn’t balance on one leg because this knee is so bad. There is no way I could stand and put all my weight on one leg, so I tried that one once and thought wow.

Trial participant 7

For others, the exercises were not a priority either because they were not motivated to do them or because they had other priorities such as being a full-time carer for a relative.

The REFORM package of care

Of the 211 participants who received an orthosis as part of the intervention and who provided a response to both questions relating to adherence to the insole and to the exercises at 12 months, 68.7% (n = 145) reported having worn their orthotic and preformed the exercises in the previous 4 weeks, 5.2% (n = 11) reported no adherence to either aspect and the remaining 55 reported that they either only wore the orthotic (n = 35) or only undertook the exercises (n = 20).

Given the multifaceted nature of the REFORM intervention, coupled with the often complex health issues of the population group, the majority of participants were able to complete only some aspects of the care package and had to adapt what was available to suit their own circumstances and perceptions of impact. A similar pragmatic approach was taken by the podiatrists, who were able to see how they could incorporate the basis of the intervention into routine practice:

I mean the majority I think I would probably give a few exercises and an insole to.

Podiatrist 7

Although podiatrists found the intervention acceptable and were, in principle, willing to administer all of the elements, some did question how the intervention could be incorporated into routine care, given the way that services were currently configured:

I think just, I suppose it just reminded me more of what I should be doing, what our roles should involve because the way we work in [centre], we’re all quite fragmented I suppose. So I would not normally have a patient in a clinic, in a routine clinic for a 20-minute appointment slot and so, if you’ve got a caseload of patients and they’re coming in to see you every 4 months, if they’ve had a fall, you’re just not going to start taking them through a falls exercise programme and there isn’t any other specialised clinic that you book them into to have that time to do it. So we have falls, so you know if somebody has a fall, our service accesses the falls clinic and the GP would refer onto the falls clinic and that’s nothing to do with podiatry.

Podiatrist 13

Some highlighted that, for the exercise element in particular, there may be more appropriate contexts in which this could be delivered:

Yeah because we have got dedicated biomechanics clinics, so I think it would sort of fit in well within that area of our service.

Podiatrist 9

Given the probable time restraints in routine care, some podiatrists suggested delivering certain elements of the package of care in group sessions to save time and money; however, opinion was split on whether or not this would be the best mode of delivery from the patients’ point of view:

Something that we could use later as maybe a group session and getting patients in to talk them through as a group session and showing them the different exercises and the type of shoe that they should be wearing.

Podiatrist 7

I think on a one-to-one basis it is better. If you have them in a class, you’ll have people who are self-conscious or not really willing to try things just out of fear and I think one-on-one situations they feel comfortable, in a safe environment and it’s a better way to deliver it.

Podiatrist 1

Footwear

Podiatrists provide footwear advice in routine practice and are well versed in doing so. The trial footwear checklist was detailed, provided a more formal evidence-based tool with which to assess footwear and was acceptable to podiatrists. For the checklist to be used in routine practice, attention would have to be paid to how the information on the checklist was recorded in the current electronic patient record systems.

Although most podiatrists found measuring for shoes straightforward, this could be time-consuming and may be difficult within the constraints of normal clinic appointments. It was suggested that this aspect could be conducted by technicians, podiatry assistant or orthotists rather than podiatrists as part of a multidisciplinary team.

It is questionable whether appropriate footwear would be provided within current NHS budgets. Outside the trial it may be more difficult to achieve adherence to footwear advice, given the financial constraints of many of the service users. In addition, there is a subgroup of service users who will be unable to access shoe retailers, who will not be able to achieve a good fit owing to existing foot problems, or who will be resistant to wearing the footwear options available.

Orthotics

Podiatrists do not routinely prescribe orthotics for falls prevention. However, they were positive about the trial orthotic and found it easy and acceptable to implement without the need for more complex biometrics assessment. Appropriate footwear is required to achieve a good fit for the orthotic; this will be less achievable for a wide range of service users outside the context of the trial if footwear is not being provided. Service users will find a pragmatic solution to incorporating wearing an orthotic, if comfortable to do so, into their everyday lives.

Exercises

Podiatrists do not currently prescribe exercise packages such as those in the REFORM intervention. They did find it acceptable to do so, in particular with the aid of the trial booklet. However, explaining the exercises properly was time-consuming and would be difficult to fit into a routine podiatry appointment. Podiatrists also felt that a follow-up appointment would be necessary to check that the exercises were being conducted appropriately to avoid injury.

The equipment necessary for the REFORM exercises was not always routinely available in podiatry clinics, meaning that additional resources would be required or that alternatives to the formal equipment would have to be suggested. Although podiatrists were happy to implement the exercise component of the intervention, this was more commonly seen as being the domain of physiotherapy or biomechanics, and in routine practice a way of incorporating the intervention into the current configuration of podiatry/falls services would have to be developed. For example, it was suggested the exercises and footwear advice could be explained in a group setting, perhaps in the context of a multidisciplinary falls clinic.