Physical activity is of vital importance to older peoples’ health. Physical activity intervention studies with older people often have low recruitment, yet little is known about non-participants.
Patients aged 60–74 years from three UK general practices were invited to participate in a nurse-supported pedometer-based walking intervention. Demographic characteristics of 298 participants and 690 non-participants were compared. Health status and physical activity of 298 participants and 183 non-participants who completed a survey were compared using age, sex adjusted odds ratios (OR) (95% confidence intervals). 15 non-participants were interviewed to explore perceived barriers to participation.
Recruitment was 30% (298/988). Participants were more likely than non-participants to be female (54% v 47%; p = 0.04) and to live in affluent postcodes (73% v 62% in top quintile; p < 0.001). Participants were more likely than non-participants who completed the survey to have an occupational pension OR 2.06 (1.35-3.13), a limiting longstanding illness OR 1.72 (1.05-2.79) and less likely to report being active OR 0.55 (0.33-0.93) or walking fast OR 0.56 (0.37-0.84). Interviewees supported general practice-based physical activity studies, particularly walking, but barriers to participation included: already sufficiently active, reluctance to walk alone or at night, physical symptoms, depression, time constraints, trial equipment and duration.
Gender and deprivation differences suggest some selection bias. However, trial participants reported more health problems and lower activity than non-participants who completed the survey, suggesting appropriate trial selection in a general practice population. Non-participant interviewees indicated that shorter interventions, addressing physical symptoms and promoting confidence in pursuing physical activity, might increase trial recruitment and uptake of practice-based physical activity endeavours.
Keywords:Physical activity; Non-participation; Primary care; Older people; Recruitment
Physical activity (PA) reduces the risk of over 20 adverse health conditions in older people, as well as improving emotional wellbeing . Physical inactivity is the fourth leading cause of death worldwide  and a major cost burden on health services .
Primary care has a key role in encouraging older people to become more active  and primary care nurses often deliver new National Health Service (NHS) Health Checks, assessing PA levels using the General Practice Physical Activity Questionnaire (GPPAQ)  and advising on increasing PA in adults up to age 74 .
Primary care PA studies report low response rates (46% for questionnaires ; 6-35% for intervention studies [7,8]). Low recruitment can lead to selection biases, thereby threatening generalisation of results. However, evidence on selection bias is contradictory: some studies reporting that trial participants are more active [9,10], and have better health  and functional status than non-participants , with other studies reporting that they have poorer health [9,12]. Such contradictions hinder the translation of findings and highlight the importance of studying trial non-participants, including a qualitative component to understand better their decision.
The PACE-Lift trial is a three month intervention designed to increase walking, using pedometers, accelerometers and one-to-one nurse consultations for older primary care patients. The target number of patients to be recruited was 300. The protocol fully describes the study design and trial interventions .
We aimed to elucidate factors influencing participation in a primary care PA trial in older adults by: 1) comparing demographic details of all those invited to participate; 2) comparing self-reported socio-demographic characteristics, health status and PA levels of participants with those of non-participants who completed a survey; and 3) exploring in interviews a sample of non-participants’ perceived barriers to participation.
A two-phase mixed-methods sequential explanatory design was used. Phase one involved collection of quantitative data using general practice records and a questionnaire survey. Phase two involved semi-structured interviews with a sample of trial non-participants. The rationale for this approach was to explain quantitative results by exploring non-participants’ views in more breadth and depth .
Three general practices in Oxfordshire and Berkshire, United Kingdom. Practices were selected that had the following: a list size >10,000 patients or >1400 patients aged 60–74 years; a practice nurse interested in carrying out the physical activity interventions; and the availability of a room for the research assistant.
Patients aged 60–74 years registered at participating practices were invited to take part in the trial if they could walk outside and had no contraindications to increasing PA. Computerised primary care records were screened by Read codes for exclusions and random samples of households were selected. (If there were 2 members of a couple living at the same address who were both potentially eligible this was a ‘double’ household; if there was only one person potentially eligible this was a ‘single’ household). General practitioners (family physicians) then scrutinised these for further exclusions, before posting study invitations (Figure 1). The invitation included trial information and a response sheet with the options of i) trial participation, ii) completing a survey or iii) no further contact. There were 298 trial participants and 690 non participants, of whom 183 completed a questionnaire.
Figure 1. PACE-Lift Trial Recruitment and Selection of Non-Participant Interviewee Sample.
This was provided by NRES Committee Oxford C 11/H0606/2.
Quantitative (Phase 1)
Postcode information recorded in computerised primary care registers allowed us to assign an index of multiple deprivation (IMD) score based on material deprivation measures, to rank individuals using national quintiles .
Table 1. Comparisons between participants and all non-participants
The odds ratios (ORs) for participating by various characteristics were estimated using logistic regression, adjusting as appropriate for possible confounding variables. Trends in ORs across IMD fifths were modelled by fitting IMD fifths as a continuous categorical variable coded 1–5. Trends across GPPAQ and General Health categories were similarly assessed, while a likelihood ratio test was used to test for variability in ORs between marital status categories.
Qualitative (Phase 2)
Permission to contact trial non-participants for a 10-minute interview was requested on the survey. Of the 183 non-participants surveyed, 77 who gave permission and 106 who did not were compared in terms of age, gender, whether invited as a couple or not, BMI and self-reported PA (GPPAQ). Fifteen of the 77 were telephoned and interviewed (Figure 1). Interviewees were purposively selected to provide: males and females of varying ages (60–74), invited to participate as an individual or a couple, samples from all three practices. Face-to-face or telephone interviews were offered and written or audio-recorded consent obtained. Recruitment stopped when no new themes were identified.
Development of interview schedule
The interviewer (AR) was a research assistant unknown to the interviewees. The semi-structured interview schedule was developed collaboratively and iteratively between AR, CV, AW and TH.
All interviews were initiated with an open question: What was your main reason for deciding not to take part in the PACE-Lift study? Probing then used previous questionnaire responses and additional questions. Possible reasons explored were: lack of time; unable/uninterested to increase PA; already physically active; not interested in research; do not want to be allocated by chance; length of programme; travel difficulties; individual versus group consultations; and trial equipment. Further trial issues were then explored: venue (e.g. general practice versus leisure centre); walking versus other PA interventions; group versus individual nurse contact; other factors that would have encouraged involvement.
The interviews were transcribed verbatim. Transcripts were subjected to an initial analysis stage ‘data reduction’  involving selecting and transforming the data as meanings and insights from the words of interviewees. AR uploaded the data onto QSR Nvivo (V9.0, Scientific Software) and read each transcript several times. This enabled identification of patterns both within and across transcripts with the aim of ensuring that recurrent codes were subject to closer scrutiny; alternative interpretations were explored, as advised in the qualitative analysis coding manual . The initial thematic framework was discussed during monthly face-to-face meetings with co-researchers who had read and coded the 15 transcripts independently, discrepancies were resolved and additional codes refined until consensus was reached. Nvivo keyword searches were used to identify overlooked material.
Comparison of participants and all non-participants using data from computerised primary care records (Table 1)
Trial recruitment was 30% (298/988). Participants were slightly but not significantly younger than non-participants (Table 1.) They were more likely to be female (p = 0.04) but being invited as a couple did not affect response. More participants lived in the most affluent quintile of Index of Multiple Deprivation, 73%, compared with 62% of non-participants (p < 0.001). However, the overall sample from the three practices was relatively affluent, with 65% (643/988) living in the least deprived quintile of deprivation, compared to 20% nationally .
Comparison of participants and non-participants completing the survey (Table 2)
Table 2. Comparisons between participants and non-participants who responded to the survey
Trial participants and non-participants completing the survey did not differ in marital or retirement status or whether they had been invited to take part as a couple. Smoking status and ethnicity showed no differences, but as only 6% were smokers and only about 1% were non-white, these comparisons lacked statistical power. However, participants were more likely to have an occupational pension, adjusted OR 2.06 (1.35-3.13). Trial participation was positively associated with having a limiting long-standing illness, adjusted OR 1.72 (1.05-2.79) and having one or more chronic diseases, adjusted OR 1.68 (1.14-2.47). Participants tended to be overweight or obese adjusted OR 1.40 (0.93-2.09) and in pain adjusted OR 1.43 (0.96-2.12), but neither difference was statistically significant. General health, use of a walking aid, falls, number of medications, mobility or self-care problems, anxiety or depression, showed no association with participation. Participation was significantly associated with lower self-reported PA: slower walking pace, adjusted OR 0.56 (0.37-0.84) and lower activity levels, adjusted OR 0.55 (0.33-0.93). The main reasons given on the survey for non-participation were: already physically active (67%); time constraints (44%); lack of interest (25%).
Qualitative interview participant selection and findings (Table 3)
Table 3. Key themes from the transcripts identifying interviewees’ reasoning behind their decision not to participate in the trial