- Email: [email protected]
Completing a postal health questionnaire did not affect anxiety or related measures: randomized controlled trial
Journal of Clinical Epidemiology 62 (2009) 74e80
Completing a postal health questionnaire did not affect anxiety or related measures: randomized controlled trial David P. Frencha,*, Helen Eborallb, Simon Griffinc, Ann Louise Kinmonthb, A. Toby Prevostb, Stephen Suttonb a Applied Research Centre in Health and Lifestyle Interventions, Coventry University, Coventry CV1 5FB General Practice and Primary Care Research Unit, Institute of Public Health, University of Cambridge, Cambridge CB2 2SR c Medical Research Council Epidemiology unit, Strangeways Research Laboratory, University of Cambridge, Cambridge CB1 8RN b
Accepted 29 February 2008
Abstract Objective: There is evidence from laboratory studies that anxiety scores are elevated by completing questionnaires about health problems for the first time. This limits interpretation of the common finding that people receiving risk information from screening programs show elevated anxiety (as assessed by questionnaire) in the short-term, which subsides over time. We examine the extent to which postal questionnaire studies are affected by this potential measurement artifact. Study Design and Setting: Participants were 964 patients at high risk of undiagnosed diabetes, registered at five general practices. A two-group randomized experimental design was used, with the experimental group (n 5 484) receiving an initial questionnaire concerning screening for diabetes, and the control group (n 5 480) not receiving this questionnaire. Outcomes were assessed by an identical questionnaire 3 months later. Results: There were no significant differences in questionnaire scores at three months between the two groups on any of the outcome measures, including anxiety, symptoms, perceptions of diabetes severity, and perceived diabetes risk. Conclusion: These results suggest that the problems associated with the use of anxiety questionnaires that are found in laboratory studies do not occur in postal studies: the observed changes in anxiety after receiving screening results are therefore unlikely to be artifactual. Ó 2008 Published by Elsevier Inc. Keywords: Questionnaire; Artifact; Bias; Epidemiology; Type 2 diabetes; Screening; Anxiety
1. Introduction An important part of the evaluation of a screening program is to estimate the extent of immediate and sustained distress brought about by that program [1]. There is good systematic review evidence that people receiving risk information from screening programs tend to show elevated anxiety and depression in the short term, which subsides over time [2]. These effects may be due to people adjusting to the information that they have received. Alternatively, given that most studies do not administer a baseline questionnaire prior to participants receiving risk information, only administering questionnaires after risk information is received, it has been argued that the pattern of distress
* Corresponding author. Applied Research Centre in Health and Lifestyle Interventions, Faculty of Health and Life Sciences, Coventry University, Whitefriars Building, Priory Street, Coventry CV1 5FB, United Kingdom. Tel.: þ44-2476-795-429; fax: þ44-2476-795-987. E-mail address: [email protected] (D.P. French). 0895-4356/09/$ e see front matter Ó 2008 Published by Elsevier Inc. doi: 10.1016/j.jclinepi.2008.02.015
observed may be at least partly due to measurement artifacts [3]. There are two measurement artifacts, which might be expected to lead to the pattern of results generally observed. First, there is evidence that completing questionnaires about health problems causes immediate elevations in anxiety. When anxiety measures are placed at the end of a questionnaire assessing beliefs about health, higher scores are obtained than when they are placed at the beginning [4(study three), 5]. Second, people appear to habituate to this initial elevated anxiety, and return to ‘‘normal’’ levels when they complete the questionnaire on subsequent occasions [4(study two)]. A study was recently published in this journal that aimed to test whether these measurement artifacts can affect the levels of anxiety and well being reported in a sample of people screened for type 2 diabetes [6]. After an initial baseline assessment, participants were randomized to receive either a single follow-up questionnaire 12 months later, or follow-up questionnaires at 1, 3, and 12 months
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
later. Response rates at 12 months (182/213 5 85.4%) were slightly lower among those who received fewer questionnaires, relative to those who also received questionnaires in the intervening time (198/218 5 90.8%), but this difference was not statistically significant (P 5 0.08). There were no significant differences between the two groups in anxiety levels on the 12-month questionnaires, although the group who received fewer questionnaires showed a significant improvement in well-being scores, relative to the group who received more questionnaires. In both groups, there was a reduction in anxiety and increase in well being observed over time [7]. In our opinion, there are two features of this study that potentially limit the validity of the conclusions that can be drawn. First, all participants were highly involved with the study before they received the experimental manipulation of questionnaires at 1 and 3 months vs. no such questionnaires. Participants were recruited via siblings with diagnosed type 2 diabetes, so consent had to be obtained from both sibling and participant, and the threat to health posed by diabetes was presumably more salient to them than to the general public. Further, participants had to attend an appointment with a nurse for a fasting plasma glucose test. As the authors themselves note, this high degree of selection and commitment required of participants is likely to have had the consequence that, ‘‘our participants may therefore have been more willing to respond to repeated questionnaires than other groups’’ [6, p. 995]. This method of selection resulted in an exceptionally high level of questionnaire response in the final sample obtained. In addition, as this method of recruitment was highly selective, those who are more anxious may not have participated. The second and more important limitation was that all participants completed anxiety and well-being measures while attending the nurse appointment at baseline. Consequently, the hypothesized effects of elevated anxiety while completing the questionnaire for the first time, and subsequent habituation, would have been missed, having already occurred prior to the experimental manipulation. The present study therefore aims to establish whether measurement artifacts can explain the apparent effects of screening on measures of psychological distress. It does this by providing an experimental test using a randomized controlled design, of the effects of completing questionnaires containing standard measures of distress and beliefs about type 2 diabetes, in a general practice sample at high risk of undiagnosed diabetes. It is particularly timely to examine potential measurement artifacts in questionnaires about screening for type 2 diabetes, given that a screening program is under consideration by the National Screening Committee [8] and that several studies examining this issue have recently been published [9]. Importantly, the present study has neither of the limitations of the study that have just been discussed [6]. Participants were invited ‘‘cold’’ from general practice registers and did not receive any invitations to screening. Further, there were no baseline questionnaires: participants were
75
randomized to either receive or not receive a questionnaire concerning diabetes, and all participants received an identical questionnaire 3 months later.
2. Methods 2.1. Participants All participants were recruited via the registers of five general practices in Cambridgeshire. These participants formed the ‘‘no screening’’ control group in a larger cluster randomized controlled trial of the effectiveness and costeffectiveness of screening for type 2 diabetes [10]. Potential participants aged 40e69 years were identified from the medical records within each practice by means of a validated algorithm based on age, gender, prescribed medication, and body mass index [11], which defined those in the top quartile of risk for having prevalent undiagnosed diabetes. Within each practice, 25% of these high-risk individuals were randomly selected for inclusion. A sample of 964 participants was thereby obtained, with the smallest practice contributing 121 participants, and the largest 263 participants. Of these 964 participants, 621 (64.4%) were male, and 343 (35.6%) were female, with a mean age of 58.6 years (SD 5 7.81). 2.2. Design A prospective two-group randomized experimental design was used, with allocation concealment and central randomization, stratified by general practice and risk score. Four hundred eighty-four participants were allocated to the experimental group receiving a baseline questionnaire concerning screening for diabetes, and 480 participants allocated to the control group not receiving such a questionnaire. Both groups were sent questionnaires identical to the experimental manipulation questionnaire at both 3 months and 12 months after the initial questionnaires were sent. The randomization and data collection process is summarized in a flowchart (Fig. 1). 2.3. Procedure At their initial contact, all participants received, without prior notification, a letter, information sheet, questionnaire, and reply-paid envelope through the post. The initial contact was immediately after randomization for the experimental group and 3 months later for the control group. The letter was from the participant’s general practice, asking for help with a study to investigate the health costs and benefits of screening for diabetes, in collaboration with the University of Cambridge. The information sheet further explained the purpose of the study as being ‘‘Your participation in this study will help to answer the important question ‘should the NHS routinely offer a screening blood test for diabetes to people in middle-age?’ An important part of this
76
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
3960 people at high risk of undiagnosed diabetes identified from medical records of 5 general practices Random 25% selected from each practice
993 people at high risk of undiagnosed diabetes 29 not included (3 died; 26 moved away)
964 participants entered into study & randomised Experimental group
Control group
484 sent t1 questionnaire
223 did not respond 261 returned t1 questionnaire
261 sent t2 questionnaire
51 did not respond
480 sent t2 questionnaire
241 did not respond
210 returned t2 questionnaire
239 returned t2 questionnaire
210 sent t3 questionnaire
239 sent t3 questionnaire
30 did not respond 180 returned t3 questionnaire
34 did not respond 205 returned t3 questionnaire
Fig. 1. Study design and recruitment flowchart.
project is to find out whether having the screening test makes people more worried or less worried about diabetes. This involves sending questionnaires to some people who are invited for screening and to others who are not invited. We are asking whether you would be willing to complete the enclosed questionnaire.’’ The information sheet also noted that, ‘‘We will ask you to complete and return the enclosed consent form and confidential questionnaire. We will then send you up to two further questionnaires over the next 12 months.’’ At no point were the participants informed that they were at high risk of undiagnosed diabetes. Importantly, to comply with ethical approval, participants who did not return questionnaires at any time point after a single reminder were not sent subsequent questionnaires. 2.4. Measures The questionnaires sent to participants at baseline, 3 months, and 12 months were identical, and consisted of the following measures in the order presented below:
Self-rated health: In line with many previous studies [12], self-rated health was assessed by a single item asking about the participant’s health in general, with the response options ‘‘excellent’’ (5), ‘‘very good’’ (4), ‘‘good’’ (3), ‘‘fair’’ (2), ‘‘poor’’ (1). Anxiety: Current levels of anxiety were assessed in adults using the six-item short form [13] of the state scale of the Spielberger State Trait Anxiety Inventory. Scores were prorated to be equivalent to those obtained with the full form of the scale, giving a range of 20e80. Anxiety was also assessed with the seven-item anxiety subscale of the Hospital Anxiety and Depression Scale [14]. Depression: Depression was assessed with the sevenitem depression subscale of the Hospital Anxiety and Depression Scale [14]. Symptoms: Participants were asked whether they had experienced each of the following 10 symptoms in the last 3 months, five being related to diabetes (d), and five not being related to diabetes (nd): ‘‘frequently feeling thirsty’’ (d); ‘‘sore throat’’ (nd); ‘‘blurred vision’’ (d); ‘‘frequent visits
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
to the toilet’’ (d); ‘‘stiff joints’’ (nd); ‘‘frequent headaches’’ (nd); ‘‘weight loss’’ (d); ‘‘breathlessness’’ (nd); ‘‘upset stomach’’ (nd); ‘‘lack of energy’’ (d). Intentions to alter lifestyle: Three items assessed intentions to (a) be more physically active, (b) eat a lower fat diet, and (c) reduce the amount of sugar in my diet, in the next 12 months, with the response options ‘‘strongly disagree’’ (1), ‘‘disagree’’ (2), neither agree nor disagree’’ (3), ‘‘agree’’ (4), ‘‘strongly agree’’ (5). Disease-specific worry: Worry about developing diabetes was assessed with a six-item adaptation of the Lerman Cancer Worry Scale [15]. Comparative risk perception: In line with many previous studies [16], perceived comparative risk of getting diabetes compared to others the participants’ age was assessed by a single item, with the response options ‘‘much lower’’ (1), ‘‘a little lower’’ (2), ‘‘about the same’’ (3), ‘‘a little higher’’ (4), ‘‘much higher’’ (5). Absolute risk perception: Perceived absolute risk of getting diabetes was assessed by a single item that has been previously used [17], asking about participants’ chances of getting diabetes at some time in their lives. There were 11 response options, in multiples of 10% from ‘‘0%’’ to ‘‘100%.’’ Perceived severity of diabetes: Perceived severity was assessed with the six-item Consequences subscale of the Revised Illness Perception Questionnaire [18]. 2.5. Analysis To examine changes in mean scores on all questionnaire measures over time, comparisons were made within the experimental group responses at baseline and at 3 months, using repeated measures t-tests. Given that only participants who returned questionnaires at an earlier time point were sent subsequent questionnaires, it was not informative to examine response rate differences over time. Chi-squared tests were used to compare frequencies of responders and nonresponders between the experimental and control group at their first contact (at baseline and 3 months, respectively) and between the two groups at 3 months. To compare mean scores on all questionnaire measures between groups, independent samples t-tests were conducted for measures completed at 3 months. As comparisons of mean scores could be biased by differential response, these analyses were repeated twice. In the first repeated analyses of mean scores at 3 months, those participants in the control group who did not return the 12-month questionnaire were removed. Thus, comparisons of mean scores at 3 months were also made for those participants who returned the first two questionnaires they were sent, that is, at baseline and 3 months for the experimental group, and at 3 months and 3 months for the control group. In the second repeated analyses of mean scores at 3 months, scale scores of people in the experimental group who did not respond at 3 months were imputed, where available, from
77
scale scores at baseline. Thus, any potential differential loss of those who were, for example, more anxious at baseline was corrected for by including their baseline anxiety scores at 3 months.
3. Results 3.1. Did questionnaire scores differ over time? The mean scores on the 11 outcome measures were compared at baseline and 3 months for those experimental group participants who returned questionnaires at both time points (see Table 1). There was one statistically significant difference between the two occasions: Participants rated their percentage risk of developing diabetes as higher on the second occasion of measurement than on the first occasion of measurement. 3.2. Do response rates differ between groups? Of the 484 people in the experimental group who were sent the baseline questionnaire, 261 (53.9%) returned it. Three months later, 210 (43.4% of the 484) responded, and at 12 months, 180 (37.2% of the 484) responded. Of the 480 people in the control group who were not sent the baseline questionnaire, 239 (49.8%) responded at 3 months, and 205 (42.7% of the 480) responded at 12 months. A chi-squared test of differences in proportions for response rates between the experimental and control group at their first contact (at baseline and 3 months, respectively) was not significant (c2 5 1.65, df 5 1, N 5 964, P 5 0.199). By contrast, a chi-squared test of differences in proportions for response rates between the experimental and control groups was significant at three months (c2 5 3.97, df 5 1, N 5 964, P 5 0.046). That is, the first time participants in both groups were asked to return questionnaires, their responses were not significantly different. However, there was a lower response rate for the experimental group on the second occasion they were sent questionnaires, compared with the control group on the first occasion they were sent questionnaires. 3.3. Do questionnaire scores differ between experimental groups? The mean scores on 11 outcome measures assessed at 3 months were compared between the two groups (see Table 2). There were no significant differences in mean scores between the two groups at this time point. Participants in the experimental group (who had therefore previously received a questionnaire) scored lower in terms of diabetes worry than participants in the control group, but this difference did not achieve statistical significance (two-tailed P 5 0.073). The effect size of this difference was small (d 5 0.18).
78
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
Table 1 Mean differences in 11 outcome measures within experimental group responders at baseline and 3 months Outcome variable
Baseline
3 months
P
Self-rated health State Trait Anxiety Inventory Hospital Anxiety and Depression Scale-anxiety Hospital Anxiety and Depression Scale-depression Symptoms (related to diabetes) Symptoms (not related to diabetes) Intentions Worry diabetes Comparative risk Percentage risk IPQ consequences
3.15 (0.85); n 5 204 31.59 (11.05); n 5 138 6.11 (4.23); n 5 200 4.19 (3.33); n 5 204 1.56 (1.25); n 5 209 1.76 (1.24); n 5 209 10.63 (2.42); n 5 209 7.76 (2.29); n 5 195 2.86 (0.94); n 5 193 29.45 (22.29); n 5 200 21.04 (3.24); n 5 198
3.16 (0.78); n 5 204 32.29 (10.81); n 5 138 5.96 (3.88); n 5 200 4.10 (3.29); n 5 204 1.51 (1.30); n 5 209 1.71 (1.27); n 5 209 10.90 (2.06); n 5 209 7.68 (2.12); n 5 195 2.84 (0.97); n 5 193 35.84 (27.25); n 5 200 21.25 (3.12); n 5 198
0.905 0.415 0.339 0.542 0.397 0.479 0.064 0.484 0.750 !0.001 0.274
These analyses on measures taken at 3 months were repeated, after removing those participants in the control group who subsequently did not return the 12-month questionnaire, to control for noncomparability of groups due to higher response attrition in the experimental group. These analyses compared the responses of similar numbers of participants in the two groups: 210/484 5 43.4% in the experimental group and 205/480 5 42.7% in the control group. In these repeated analyses, there were still no significant differences in mean scores, although participants in the experimental group again scored lower in terms of diabetes worry than participants in the control group (two-tailed P 5 0.063). The comparisons of measures taken at 3 months between groups were repeated again, with scale scores of people in the experimental group who did not respond at 3 months being imputed, where available, from scale scores at baseline. In these repeated analyses, there were again no significant differences in mean scores, although participants in the experimental group scored higher on state anxiety (as assessed by the State Trait Anxiety Inventory) than participants in the control group (two-tailed P 5 0.063).
4. Discussion Mean scores on only one of the 11 outcome measures (perceived absolute risk of developing diabetes) changed
between baseline and 3 months for experimental group participants. Responses rates at first contact were similar for the two groups, but participants in the experimental group (who completed a questionnaire at baseline) were slightly less likely to complete questionnaires at 3 months, compared to participants in the control group (who were not sent a questionnaire at baseline). There were no significant differences in questionnaire scores at 3 months between the two groups on any outcome measures, including anxiety, symptoms, diabetes severity, and comparative or absolute risk. This lack of significant differences in questionnaire scores at 3 months between the two groups remained even when the differences in responding were controlled for using two different methods. It has previously been observed that when people complete questionnaires about health threats for the first time, they score higher on measures of anxiety than when they subsequently complete the same questionnaire [4]. In the present study, only one of the 11 outcome measures (perceived absolute risk of developing diabetes) differed between baseline and 3 months. We attribute this single significant change to chance: there is no reason to expect scores on this measure to change over time, but not change on other measures, including perceived comparative risk of developing diabetes. More importantly, there were no significant differences in the mean questionnaire scores at 3 months for participants who completed baseline questionnaires concerning diabetes, and those who were not sent such questionnaires.
Table 2 Mean differences in 11 outcome measures between the experimental and control group responders at 3 months Outcome variable
Experimental group
Control group
P
Self-rated health State Trait Anxiety Inventory (a 5 0.84) Hospital Anxiety and Depression Scale-anxiety (a 5 0.68) Hospital Anxiety and Depression Scale-depression (a 5 0.62) Symptoms (related to diabetes) Symptoms (not related to diabetes) Intentions (a 5 0.79) Worry diabetes (a 5 0.81) Comparative risk Percentage risk IPQ consequences (a 5 0.72)
3.15 (0.78); n 5 210 32.52 (11.15); n 5 164 5.93 (3.85); n 5 205 4.14 (3.32); n 5 206 1.51 (1.30); n 5 209 1.71 (1.27); n 5 209 10.90 (2.06); n 5 209 7.65 (2.11); n 5 199 2.83 (0.97); n 5 197 36.53 (27.90); n 5 209 21.25 (3.16); n 5 204
3.12 (0.82); n 5 233 31.16 (11.57); n 5 194 6.01 (3.88); n 5 237 4.21 (3.44); n 5 238 1.51 (1.28); n 5 239 1.88 (1.30); n 5 239 10.84 (2.20); n 5 237 8.06 (2.53); n 5 229 2.88 (0.91); n 5 229 35.97 (25.22); n 5 239 21.32 (3.58); n 5 237
0.715 0.260 0.825 0.830 0.951 0.161 0.734 0.073 0.588 0.824 0.816
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
Overall, the results of the present study indicate that completing a postal questionnaire about health risk does not lead to any discernible elevations in anxiety, depression, or any of several other outcome measures examined. These results are therefore consistent with those of a similar study recently published in this journal [6], although as noted above, we believe that the design of the present study allows greater confidence in this conclusion than the earlier study. We believe there are three plausible explanations for the discrepancy between our results and those of previous laboratory-based research [5, 6]. First, the designs differed. In the laboratory-based research, a comparison was made between the scores obtained from people who completed anxiety scales at the beginning of a questionnaire with people who completed anxiety scales at the end of a questionnaire. In the present study, the comparison were made between people who completed the same questionnaire at an interval of 3 months, and between people who completed the same questionnaire, one group of which had completed the same questionnaire 3 months earlier. Similarly, in the previous research, due to the face-to-face method of data collection, participants were required to complete the questionnaires immediately upon receiving them, whereas the present research involved postal questionnaires. Consequently, participants in the present study may not have completed the questionnaires immediately, instead completing them after any short-term elevations in experienced anxiety passed. By design, our study therefore examined effects over a much longer time period, and one might expect that any immediate increases in anxiety would dissipate over time. Second, in the previous research, there was minimal dropout, whereas in the present research it was around 50%e60%. Given that people who are more anxious are less likely to return questionnaires about health [19, 20], it may be that nonresponders in the present study were those who experienced elevated anxiety initially, and who would therefore show a reduction in anxiety upon completion of questionnaires for the second time. It is generally observed that people receiving risk information from screening programs tend to show elevated anxiety in the short-term, which subsides over time [2]. The results of the present study suggest that it is unlikely that these effects are artifactual, resulting from being asked to complete questionnaires. In contrast to the laboratory studies, there was little evidence that participants in the present study reported higher anxiety scores on the first occasion they completed questionnaires about their health, in either within-subjects or between-subjects comparisons. Consequently, it appears reasonable to take the results of studies on receiving risk information from screening programs at face value: they are due to people adjusting to the information they have received. There are two caveats to this conclusion. First, participants in the experimental group (who had therefore previously received a questionnaire) scored slightly lower in
79
terms of diabetes worry than participants in the control group. This difference was not statistically significant, and the effect size was small (d 5 0.18). Nevertheless, the present study cannot rule out the possibility that completing questionnaires may lead to some distress which has not completely disappeared when completing a subsequent questionnaire months later. The results of the present study suggest that these effects are likely to be small at most, and therefore not of sufficient size to be responsible for the patterns of distress observed in response to receiving risk information from screening programs. The second caveat concerns the difference in response rates observed at 3 months between the experimental and control groups, with participants in the control group being more likely to respond at 3 months. People who are more anxious in the aftermath of screening are less likely to return questionnaires [19, 20]. It is possible in the present study that the more anxious participants in the control group may have been less likely to return questionnaires at three months, leading to apparently similar levels of anxiety as those participants in the experimental group who had overcome any anxiety evoked as a result of completing baseline questionnaires 3 months earlier. We believe that this explanation is unlikely. To examine this possibility, two further analyses of 3-month questionnaires were conducted. First, only the responses of participants who returned both their first two questionnaires were included (at baseline and 3 months in the experimental group, and at 3 months and 12 months in the control group). Second, where there was nonresponse from participants in the experimental group at 3 months, scores were imputed from baseline. In these repeated analyses, there were still no significant differences in mean questionnaire scores on all outcomes at 3 months. We believe the results of the present study have provided good evidence that the patterns of distress observed in questionnaire responses from people receiving risk information from screening programs are not due to the reactive effects of completing questionnaires, but instead reflect genuine patterns of emotional responses. Nevertheless, we believe that there is still a need for greater understanding of the reactive effects of completing questionnaires about threats to health. Studies conducted using much shorter time frames than the present study have found that completing questionnaires about health problems causes elevations in anxiety [4(study three), 5], and that people return to ‘‘normal’’ levels when they complete the questionnaire on subsequent occasions [4(study two)]. These effects can be expected to occur when studies involve people completing questionnaires immediately after receiving medical interventions, rather than completing questionnaires by post, as in the present research. It is still necessary to quantify the circumstances under which these effects can bias epidemiological studies examining distress due to medical interventions. Specifically, it is not clear over what period these effects can last, the
80
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
extent to which mode of administration (e.g., postal questionnaire vs. face-to-face administration) is important in producing these effects, which people are most susceptible to these potential reactive effects of questionnaires, and which measures may be most susceptible. Future research should aim to disentangle the effects of each of the variables just mentioned, to understand the processes that lead to these effects, and thereby reduce the chances of these biases leading to misinterpretation of questionnaires about anxiety and well being in future research.
Acknowledgments This study was funded by project grant from The Wellcome Trust (reference number 071200/Z/03/Z). NHS R&D provided support for science funding to cover the NHS costs of hosting the research. We thank the practices and patients who participated in this study. References [1] National Screening Committee. First report of the National Screening Committee. London: Health Departments of the United Kingdom; 1998. Available at. www.nsc.nhs.uk/pdfs/nsc_firstreport.pdf. Accessed May 1, 2008. [2] Shaw C, Abrams K, Marteau TM. Psychological impact of predicting individuals’ risks of illness: a systematic review. Soc Sci Med 1999;49:1571e98. [3] Johnston M, French DP, Bonetti D, Johnston DW. Assessment and measurement in health psychology. In: Sutton S, Baum A, Johnston M, editors. Handbook of health psychology. London: Sage; 2004. p. 288e323. [4] Johnston M. Mood in chronic disease: questioning the answers. Curr Psychol 1999;18:71e87. [5] Lister AM, Rode S, Farmer A, Salkovskis P. Does thinking about personal health risk increase anxiety? J Health Psychol 2002;7:409e14. [6] Farmer AJ, Doll HA. In a randomized trial, outcomes were not affected by intensive follow-up over 1 year. J Clini Epidemiol 2005;58:991e6.
[7] Farmer AJ, Doll H, Levy JC, Salkovskis PM. The impact of screening for Type 2 diabetes in siblings of patients with established diabetes. Diabetic Med 2003;20:996e1004. [8] Waugh N, Scotland G, McNamee P, Gillett M, Brennan A, Goyder E, et al. Screening for type 2 diabetes: literature review and economic modelling. Health Technol Assess 2007;11:17. [9] Adriaanse MC, Snoek FJ. The psychological impact of screening for type 2 diabetes. Diabetes Metab Res Rev 2006;22:20e5. [10] Eborall HC, Griffin SJ, Prevost AT, Kinmonth AL, French DP, Sutton S. Psychological impact of screening for type 2 diabetes: controlled trial and comparative study embedded in the ADDITION (Cambridge) randomised controlled trial. BMJ 2007;335(7619): 486e9. [11] Griffin SJ, Little PS, Kinmonth AL, Hales CN, Wareham NJ. Diabetes risk score: towards earlier detection of type 2 diabetes in general practice. Diabetes Metab Res Rev 2000;16:164e71. [12] Idler EL, Benyamini Y. Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behav 2007;38: 21e37. [13] Marteau TM, Bekker H. The development of a six-item short-form of the state scale of the Spielberger state-trait anxiety inventory (STAI). Br J Clin Psychol 1992;31:301e6. [14] Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand 1983;67:361e70. [15] Lerman C, Trock B, Rimer BK, Jepson C, Brody D, Boyce A. Psychological side effects of breast cancer screening. Health Psychol 1991;10:259e67. [16] Weinstein ND. Perceptions of personal susceptibility to harm. In: Mays VM, Albee GW, Schneider SF, editors. Primary prevention of AIDS: Psychological approaches. Newbury Park, CA: Sage; 1989. p. 142e67. [17] Weinstein ND, Diefenbach MA. Percentage and verbal category measures of risk likelihood. Health Educ Res 1997;12:139e41. [18] Moss-Morris R, Weinman J, Petrie KJ, Horne R, Cameron LD, Buick D. The revised illness perception questionnaire (IPQ-R). Psychol Health 2002;17:1e16. [19] Timman R, Roos R, Maat-Kievit A, Tibben A. Adverse effects of predictive testing for Huntington disease underestimated: long-term effects 7-10 years after the test. Health Psychol 2004;23:189e97. [20] Maissi E, Marteau TM, Hankins M, Moss S, Legood R, Gray A. The psychological impact of human papillomavirus testing in women with borderline or mildly dyskaryotic cervical smear test results: 6-month follow up. Br J Cancer 2005;92:990e4.
Completing a postal health questionnaire did not affect anxiety or related measures: randomized controlled trial David P. Frencha,*, Helen Eborallb, Simon Griffinc, Ann Louise Kinmonthb, A. Toby Prevostb, Stephen Suttonb a Applied Research Centre in Health and Lifestyle Interventions, Coventry University, Coventry CV1 5FB General Practice and Primary Care Research Unit, Institute of Public Health, University of Cambridge, Cambridge CB2 2SR c Medical Research Council Epidemiology unit, Strangeways Research Laboratory, University of Cambridge, Cambridge CB1 8RN b
Accepted 29 February 2008
Abstract Objective: There is evidence from laboratory studies that anxiety scores are elevated by completing questionnaires about health problems for the first time. This limits interpretation of the common finding that people receiving risk information from screening programs show elevated anxiety (as assessed by questionnaire) in the short-term, which subsides over time. We examine the extent to which postal questionnaire studies are affected by this potential measurement artifact. Study Design and Setting: Participants were 964 patients at high risk of undiagnosed diabetes, registered at five general practices. A two-group randomized experimental design was used, with the experimental group (n 5 484) receiving an initial questionnaire concerning screening for diabetes, and the control group (n 5 480) not receiving this questionnaire. Outcomes were assessed by an identical questionnaire 3 months later. Results: There were no significant differences in questionnaire scores at three months between the two groups on any of the outcome measures, including anxiety, symptoms, perceptions of diabetes severity, and perceived diabetes risk. Conclusion: These results suggest that the problems associated with the use of anxiety questionnaires that are found in laboratory studies do not occur in postal studies: the observed changes in anxiety after receiving screening results are therefore unlikely to be artifactual. Ó 2008 Published by Elsevier Inc. Keywords: Questionnaire; Artifact; Bias; Epidemiology; Type 2 diabetes; Screening; Anxiety
1. Introduction An important part of the evaluation of a screening program is to estimate the extent of immediate and sustained distress brought about by that program [1]. There is good systematic review evidence that people receiving risk information from screening programs tend to show elevated anxiety and depression in the short term, which subsides over time [2]. These effects may be due to people adjusting to the information that they have received. Alternatively, given that most studies do not administer a baseline questionnaire prior to participants receiving risk information, only administering questionnaires after risk information is received, it has been argued that the pattern of distress
* Corresponding author. Applied Research Centre in Health and Lifestyle Interventions, Faculty of Health and Life Sciences, Coventry University, Whitefriars Building, Priory Street, Coventry CV1 5FB, United Kingdom. Tel.: þ44-2476-795-429; fax: þ44-2476-795-987. E-mail address: [email protected] (D.P. French). 0895-4356/09/$ e see front matter Ó 2008 Published by Elsevier Inc. doi: 10.1016/j.jclinepi.2008.02.015
observed may be at least partly due to measurement artifacts [3]. There are two measurement artifacts, which might be expected to lead to the pattern of results generally observed. First, there is evidence that completing questionnaires about health problems causes immediate elevations in anxiety. When anxiety measures are placed at the end of a questionnaire assessing beliefs about health, higher scores are obtained than when they are placed at the beginning [4(study three), 5]. Second, people appear to habituate to this initial elevated anxiety, and return to ‘‘normal’’ levels when they complete the questionnaire on subsequent occasions [4(study two)]. A study was recently published in this journal that aimed to test whether these measurement artifacts can affect the levels of anxiety and well being reported in a sample of people screened for type 2 diabetes [6]. After an initial baseline assessment, participants were randomized to receive either a single follow-up questionnaire 12 months later, or follow-up questionnaires at 1, 3, and 12 months
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
later. Response rates at 12 months (182/213 5 85.4%) were slightly lower among those who received fewer questionnaires, relative to those who also received questionnaires in the intervening time (198/218 5 90.8%), but this difference was not statistically significant (P 5 0.08). There were no significant differences between the two groups in anxiety levels on the 12-month questionnaires, although the group who received fewer questionnaires showed a significant improvement in well-being scores, relative to the group who received more questionnaires. In both groups, there was a reduction in anxiety and increase in well being observed over time [7]. In our opinion, there are two features of this study that potentially limit the validity of the conclusions that can be drawn. First, all participants were highly involved with the study before they received the experimental manipulation of questionnaires at 1 and 3 months vs. no such questionnaires. Participants were recruited via siblings with diagnosed type 2 diabetes, so consent had to be obtained from both sibling and participant, and the threat to health posed by diabetes was presumably more salient to them than to the general public. Further, participants had to attend an appointment with a nurse for a fasting plasma glucose test. As the authors themselves note, this high degree of selection and commitment required of participants is likely to have had the consequence that, ‘‘our participants may therefore have been more willing to respond to repeated questionnaires than other groups’’ [6, p. 995]. This method of selection resulted in an exceptionally high level of questionnaire response in the final sample obtained. In addition, as this method of recruitment was highly selective, those who are more anxious may not have participated. The second and more important limitation was that all participants completed anxiety and well-being measures while attending the nurse appointment at baseline. Consequently, the hypothesized effects of elevated anxiety while completing the questionnaire for the first time, and subsequent habituation, would have been missed, having already occurred prior to the experimental manipulation. The present study therefore aims to establish whether measurement artifacts can explain the apparent effects of screening on measures of psychological distress. It does this by providing an experimental test using a randomized controlled design, of the effects of completing questionnaires containing standard measures of distress and beliefs about type 2 diabetes, in a general practice sample at high risk of undiagnosed diabetes. It is particularly timely to examine potential measurement artifacts in questionnaires about screening for type 2 diabetes, given that a screening program is under consideration by the National Screening Committee [8] and that several studies examining this issue have recently been published [9]. Importantly, the present study has neither of the limitations of the study that have just been discussed [6]. Participants were invited ‘‘cold’’ from general practice registers and did not receive any invitations to screening. Further, there were no baseline questionnaires: participants were
75
randomized to either receive or not receive a questionnaire concerning diabetes, and all participants received an identical questionnaire 3 months later.
2. Methods 2.1. Participants All participants were recruited via the registers of five general practices in Cambridgeshire. These participants formed the ‘‘no screening’’ control group in a larger cluster randomized controlled trial of the effectiveness and costeffectiveness of screening for type 2 diabetes [10]. Potential participants aged 40e69 years were identified from the medical records within each practice by means of a validated algorithm based on age, gender, prescribed medication, and body mass index [11], which defined those in the top quartile of risk for having prevalent undiagnosed diabetes. Within each practice, 25% of these high-risk individuals were randomly selected for inclusion. A sample of 964 participants was thereby obtained, with the smallest practice contributing 121 participants, and the largest 263 participants. Of these 964 participants, 621 (64.4%) were male, and 343 (35.6%) were female, with a mean age of 58.6 years (SD 5 7.81). 2.2. Design A prospective two-group randomized experimental design was used, with allocation concealment and central randomization, stratified by general practice and risk score. Four hundred eighty-four participants were allocated to the experimental group receiving a baseline questionnaire concerning screening for diabetes, and 480 participants allocated to the control group not receiving such a questionnaire. Both groups were sent questionnaires identical to the experimental manipulation questionnaire at both 3 months and 12 months after the initial questionnaires were sent. The randomization and data collection process is summarized in a flowchart (Fig. 1). 2.3. Procedure At their initial contact, all participants received, without prior notification, a letter, information sheet, questionnaire, and reply-paid envelope through the post. The initial contact was immediately after randomization for the experimental group and 3 months later for the control group. The letter was from the participant’s general practice, asking for help with a study to investigate the health costs and benefits of screening for diabetes, in collaboration with the University of Cambridge. The information sheet further explained the purpose of the study as being ‘‘Your participation in this study will help to answer the important question ‘should the NHS routinely offer a screening blood test for diabetes to people in middle-age?’ An important part of this
76
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
3960 people at high risk of undiagnosed diabetes identified from medical records of 5 general practices Random 25% selected from each practice
993 people at high risk of undiagnosed diabetes 29 not included (3 died; 26 moved away)
964 participants entered into study & randomised Experimental group
Control group
484 sent t1 questionnaire
223 did not respond 261 returned t1 questionnaire
261 sent t2 questionnaire
51 did not respond
480 sent t2 questionnaire
241 did not respond
210 returned t2 questionnaire
239 returned t2 questionnaire
210 sent t3 questionnaire
239 sent t3 questionnaire
30 did not respond 180 returned t3 questionnaire
34 did not respond 205 returned t3 questionnaire
Fig. 1. Study design and recruitment flowchart.
project is to find out whether having the screening test makes people more worried or less worried about diabetes. This involves sending questionnaires to some people who are invited for screening and to others who are not invited. We are asking whether you would be willing to complete the enclosed questionnaire.’’ The information sheet also noted that, ‘‘We will ask you to complete and return the enclosed consent form and confidential questionnaire. We will then send you up to two further questionnaires over the next 12 months.’’ At no point were the participants informed that they were at high risk of undiagnosed diabetes. Importantly, to comply with ethical approval, participants who did not return questionnaires at any time point after a single reminder were not sent subsequent questionnaires. 2.4. Measures The questionnaires sent to participants at baseline, 3 months, and 12 months were identical, and consisted of the following measures in the order presented below:
Self-rated health: In line with many previous studies [12], self-rated health was assessed by a single item asking about the participant’s health in general, with the response options ‘‘excellent’’ (5), ‘‘very good’’ (4), ‘‘good’’ (3), ‘‘fair’’ (2), ‘‘poor’’ (1). Anxiety: Current levels of anxiety were assessed in adults using the six-item short form [13] of the state scale of the Spielberger State Trait Anxiety Inventory. Scores were prorated to be equivalent to those obtained with the full form of the scale, giving a range of 20e80. Anxiety was also assessed with the seven-item anxiety subscale of the Hospital Anxiety and Depression Scale [14]. Depression: Depression was assessed with the sevenitem depression subscale of the Hospital Anxiety and Depression Scale [14]. Symptoms: Participants were asked whether they had experienced each of the following 10 symptoms in the last 3 months, five being related to diabetes (d), and five not being related to diabetes (nd): ‘‘frequently feeling thirsty’’ (d); ‘‘sore throat’’ (nd); ‘‘blurred vision’’ (d); ‘‘frequent visits
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
to the toilet’’ (d); ‘‘stiff joints’’ (nd); ‘‘frequent headaches’’ (nd); ‘‘weight loss’’ (d); ‘‘breathlessness’’ (nd); ‘‘upset stomach’’ (nd); ‘‘lack of energy’’ (d). Intentions to alter lifestyle: Three items assessed intentions to (a) be more physically active, (b) eat a lower fat diet, and (c) reduce the amount of sugar in my diet, in the next 12 months, with the response options ‘‘strongly disagree’’ (1), ‘‘disagree’’ (2), neither agree nor disagree’’ (3), ‘‘agree’’ (4), ‘‘strongly agree’’ (5). Disease-specific worry: Worry about developing diabetes was assessed with a six-item adaptation of the Lerman Cancer Worry Scale [15]. Comparative risk perception: In line with many previous studies [16], perceived comparative risk of getting diabetes compared to others the participants’ age was assessed by a single item, with the response options ‘‘much lower’’ (1), ‘‘a little lower’’ (2), ‘‘about the same’’ (3), ‘‘a little higher’’ (4), ‘‘much higher’’ (5). Absolute risk perception: Perceived absolute risk of getting diabetes was assessed by a single item that has been previously used [17], asking about participants’ chances of getting diabetes at some time in their lives. There were 11 response options, in multiples of 10% from ‘‘0%’’ to ‘‘100%.’’ Perceived severity of diabetes: Perceived severity was assessed with the six-item Consequences subscale of the Revised Illness Perception Questionnaire [18]. 2.5. Analysis To examine changes in mean scores on all questionnaire measures over time, comparisons were made within the experimental group responses at baseline and at 3 months, using repeated measures t-tests. Given that only participants who returned questionnaires at an earlier time point were sent subsequent questionnaires, it was not informative to examine response rate differences over time. Chi-squared tests were used to compare frequencies of responders and nonresponders between the experimental and control group at their first contact (at baseline and 3 months, respectively) and between the two groups at 3 months. To compare mean scores on all questionnaire measures between groups, independent samples t-tests were conducted for measures completed at 3 months. As comparisons of mean scores could be biased by differential response, these analyses were repeated twice. In the first repeated analyses of mean scores at 3 months, those participants in the control group who did not return the 12-month questionnaire were removed. Thus, comparisons of mean scores at 3 months were also made for those participants who returned the first two questionnaires they were sent, that is, at baseline and 3 months for the experimental group, and at 3 months and 3 months for the control group. In the second repeated analyses of mean scores at 3 months, scale scores of people in the experimental group who did not respond at 3 months were imputed, where available, from
77
scale scores at baseline. Thus, any potential differential loss of those who were, for example, more anxious at baseline was corrected for by including their baseline anxiety scores at 3 months.
3. Results 3.1. Did questionnaire scores differ over time? The mean scores on the 11 outcome measures were compared at baseline and 3 months for those experimental group participants who returned questionnaires at both time points (see Table 1). There was one statistically significant difference between the two occasions: Participants rated their percentage risk of developing diabetes as higher on the second occasion of measurement than on the first occasion of measurement. 3.2. Do response rates differ between groups? Of the 484 people in the experimental group who were sent the baseline questionnaire, 261 (53.9%) returned it. Three months later, 210 (43.4% of the 484) responded, and at 12 months, 180 (37.2% of the 484) responded. Of the 480 people in the control group who were not sent the baseline questionnaire, 239 (49.8%) responded at 3 months, and 205 (42.7% of the 480) responded at 12 months. A chi-squared test of differences in proportions for response rates between the experimental and control group at their first contact (at baseline and 3 months, respectively) was not significant (c2 5 1.65, df 5 1, N 5 964, P 5 0.199). By contrast, a chi-squared test of differences in proportions for response rates between the experimental and control groups was significant at three months (c2 5 3.97, df 5 1, N 5 964, P 5 0.046). That is, the first time participants in both groups were asked to return questionnaires, their responses were not significantly different. However, there was a lower response rate for the experimental group on the second occasion they were sent questionnaires, compared with the control group on the first occasion they were sent questionnaires. 3.3. Do questionnaire scores differ between experimental groups? The mean scores on 11 outcome measures assessed at 3 months were compared between the two groups (see Table 2). There were no significant differences in mean scores between the two groups at this time point. Participants in the experimental group (who had therefore previously received a questionnaire) scored lower in terms of diabetes worry than participants in the control group, but this difference did not achieve statistical significance (two-tailed P 5 0.073). The effect size of this difference was small (d 5 0.18).
78
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
Table 1 Mean differences in 11 outcome measures within experimental group responders at baseline and 3 months Outcome variable
Baseline
3 months
P
Self-rated health State Trait Anxiety Inventory Hospital Anxiety and Depression Scale-anxiety Hospital Anxiety and Depression Scale-depression Symptoms (related to diabetes) Symptoms (not related to diabetes) Intentions Worry diabetes Comparative risk Percentage risk IPQ consequences
3.15 (0.85); n 5 204 31.59 (11.05); n 5 138 6.11 (4.23); n 5 200 4.19 (3.33); n 5 204 1.56 (1.25); n 5 209 1.76 (1.24); n 5 209 10.63 (2.42); n 5 209 7.76 (2.29); n 5 195 2.86 (0.94); n 5 193 29.45 (22.29); n 5 200 21.04 (3.24); n 5 198
3.16 (0.78); n 5 204 32.29 (10.81); n 5 138 5.96 (3.88); n 5 200 4.10 (3.29); n 5 204 1.51 (1.30); n 5 209 1.71 (1.27); n 5 209 10.90 (2.06); n 5 209 7.68 (2.12); n 5 195 2.84 (0.97); n 5 193 35.84 (27.25); n 5 200 21.25 (3.12); n 5 198
0.905 0.415 0.339 0.542 0.397 0.479 0.064 0.484 0.750 !0.001 0.274
These analyses on measures taken at 3 months were repeated, after removing those participants in the control group who subsequently did not return the 12-month questionnaire, to control for noncomparability of groups due to higher response attrition in the experimental group. These analyses compared the responses of similar numbers of participants in the two groups: 210/484 5 43.4% in the experimental group and 205/480 5 42.7% in the control group. In these repeated analyses, there were still no significant differences in mean scores, although participants in the experimental group again scored lower in terms of diabetes worry than participants in the control group (two-tailed P 5 0.063). The comparisons of measures taken at 3 months between groups were repeated again, with scale scores of people in the experimental group who did not respond at 3 months being imputed, where available, from scale scores at baseline. In these repeated analyses, there were again no significant differences in mean scores, although participants in the experimental group scored higher on state anxiety (as assessed by the State Trait Anxiety Inventory) than participants in the control group (two-tailed P 5 0.063).
4. Discussion Mean scores on only one of the 11 outcome measures (perceived absolute risk of developing diabetes) changed
between baseline and 3 months for experimental group participants. Responses rates at first contact were similar for the two groups, but participants in the experimental group (who completed a questionnaire at baseline) were slightly less likely to complete questionnaires at 3 months, compared to participants in the control group (who were not sent a questionnaire at baseline). There were no significant differences in questionnaire scores at 3 months between the two groups on any outcome measures, including anxiety, symptoms, diabetes severity, and comparative or absolute risk. This lack of significant differences in questionnaire scores at 3 months between the two groups remained even when the differences in responding were controlled for using two different methods. It has previously been observed that when people complete questionnaires about health threats for the first time, they score higher on measures of anxiety than when they subsequently complete the same questionnaire [4]. In the present study, only one of the 11 outcome measures (perceived absolute risk of developing diabetes) differed between baseline and 3 months. We attribute this single significant change to chance: there is no reason to expect scores on this measure to change over time, but not change on other measures, including perceived comparative risk of developing diabetes. More importantly, there were no significant differences in the mean questionnaire scores at 3 months for participants who completed baseline questionnaires concerning diabetes, and those who were not sent such questionnaires.
Table 2 Mean differences in 11 outcome measures between the experimental and control group responders at 3 months Outcome variable
Experimental group
Control group
P
Self-rated health State Trait Anxiety Inventory (a 5 0.84) Hospital Anxiety and Depression Scale-anxiety (a 5 0.68) Hospital Anxiety and Depression Scale-depression (a 5 0.62) Symptoms (related to diabetes) Symptoms (not related to diabetes) Intentions (a 5 0.79) Worry diabetes (a 5 0.81) Comparative risk Percentage risk IPQ consequences (a 5 0.72)
3.15 (0.78); n 5 210 32.52 (11.15); n 5 164 5.93 (3.85); n 5 205 4.14 (3.32); n 5 206 1.51 (1.30); n 5 209 1.71 (1.27); n 5 209 10.90 (2.06); n 5 209 7.65 (2.11); n 5 199 2.83 (0.97); n 5 197 36.53 (27.90); n 5 209 21.25 (3.16); n 5 204
3.12 (0.82); n 5 233 31.16 (11.57); n 5 194 6.01 (3.88); n 5 237 4.21 (3.44); n 5 238 1.51 (1.28); n 5 239 1.88 (1.30); n 5 239 10.84 (2.20); n 5 237 8.06 (2.53); n 5 229 2.88 (0.91); n 5 229 35.97 (25.22); n 5 239 21.32 (3.58); n 5 237
0.715 0.260 0.825 0.830 0.951 0.161 0.734 0.073 0.588 0.824 0.816
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
Overall, the results of the present study indicate that completing a postal questionnaire about health risk does not lead to any discernible elevations in anxiety, depression, or any of several other outcome measures examined. These results are therefore consistent with those of a similar study recently published in this journal [6], although as noted above, we believe that the design of the present study allows greater confidence in this conclusion than the earlier study. We believe there are three plausible explanations for the discrepancy between our results and those of previous laboratory-based research [5, 6]. First, the designs differed. In the laboratory-based research, a comparison was made between the scores obtained from people who completed anxiety scales at the beginning of a questionnaire with people who completed anxiety scales at the end of a questionnaire. In the present study, the comparison were made between people who completed the same questionnaire at an interval of 3 months, and between people who completed the same questionnaire, one group of which had completed the same questionnaire 3 months earlier. Similarly, in the previous research, due to the face-to-face method of data collection, participants were required to complete the questionnaires immediately upon receiving them, whereas the present research involved postal questionnaires. Consequently, participants in the present study may not have completed the questionnaires immediately, instead completing them after any short-term elevations in experienced anxiety passed. By design, our study therefore examined effects over a much longer time period, and one might expect that any immediate increases in anxiety would dissipate over time. Second, in the previous research, there was minimal dropout, whereas in the present research it was around 50%e60%. Given that people who are more anxious are less likely to return questionnaires about health [19, 20], it may be that nonresponders in the present study were those who experienced elevated anxiety initially, and who would therefore show a reduction in anxiety upon completion of questionnaires for the second time. It is generally observed that people receiving risk information from screening programs tend to show elevated anxiety in the short-term, which subsides over time [2]. The results of the present study suggest that it is unlikely that these effects are artifactual, resulting from being asked to complete questionnaires. In contrast to the laboratory studies, there was little evidence that participants in the present study reported higher anxiety scores on the first occasion they completed questionnaires about their health, in either within-subjects or between-subjects comparisons. Consequently, it appears reasonable to take the results of studies on receiving risk information from screening programs at face value: they are due to people adjusting to the information they have received. There are two caveats to this conclusion. First, participants in the experimental group (who had therefore previously received a questionnaire) scored slightly lower in
79
terms of diabetes worry than participants in the control group. This difference was not statistically significant, and the effect size was small (d 5 0.18). Nevertheless, the present study cannot rule out the possibility that completing questionnaires may lead to some distress which has not completely disappeared when completing a subsequent questionnaire months later. The results of the present study suggest that these effects are likely to be small at most, and therefore not of sufficient size to be responsible for the patterns of distress observed in response to receiving risk information from screening programs. The second caveat concerns the difference in response rates observed at 3 months between the experimental and control groups, with participants in the control group being more likely to respond at 3 months. People who are more anxious in the aftermath of screening are less likely to return questionnaires [19, 20]. It is possible in the present study that the more anxious participants in the control group may have been less likely to return questionnaires at three months, leading to apparently similar levels of anxiety as those participants in the experimental group who had overcome any anxiety evoked as a result of completing baseline questionnaires 3 months earlier. We believe that this explanation is unlikely. To examine this possibility, two further analyses of 3-month questionnaires were conducted. First, only the responses of participants who returned both their first two questionnaires were included (at baseline and 3 months in the experimental group, and at 3 months and 12 months in the control group). Second, where there was nonresponse from participants in the experimental group at 3 months, scores were imputed from baseline. In these repeated analyses, there were still no significant differences in mean questionnaire scores on all outcomes at 3 months. We believe the results of the present study have provided good evidence that the patterns of distress observed in questionnaire responses from people receiving risk information from screening programs are not due to the reactive effects of completing questionnaires, but instead reflect genuine patterns of emotional responses. Nevertheless, we believe that there is still a need for greater understanding of the reactive effects of completing questionnaires about threats to health. Studies conducted using much shorter time frames than the present study have found that completing questionnaires about health problems causes elevations in anxiety [4(study three), 5], and that people return to ‘‘normal’’ levels when they complete the questionnaire on subsequent occasions [4(study two)]. These effects can be expected to occur when studies involve people completing questionnaires immediately after receiving medical interventions, rather than completing questionnaires by post, as in the present research. It is still necessary to quantify the circumstances under which these effects can bias epidemiological studies examining distress due to medical interventions. Specifically, it is not clear over what period these effects can last, the
80
D.P. French et al. / Journal of Clinical Epidemiology 62 (2009) 74e80
extent to which mode of administration (e.g., postal questionnaire vs. face-to-face administration) is important in producing these effects, which people are most susceptible to these potential reactive effects of questionnaires, and which measures may be most susceptible. Future research should aim to disentangle the effects of each of the variables just mentioned, to understand the processes that lead to these effects, and thereby reduce the chances of these biases leading to misinterpretation of questionnaires about anxiety and well being in future research.
Acknowledgments This study was funded by project grant from The Wellcome Trust (reference number 071200/Z/03/Z). NHS R&D provided support for science funding to cover the NHS costs of hosting the research. We thank the practices and patients who participated in this study. References [1] National Screening Committee. First report of the National Screening Committee. London: Health Departments of the United Kingdom; 1998. Available at. www.nsc.nhs.uk/pdfs/nsc_firstreport.pdf. Accessed May 1, 2008. [2] Shaw C, Abrams K, Marteau TM. Psychological impact of predicting individuals’ risks of illness: a systematic review. Soc Sci Med 1999;49:1571e98. [3] Johnston M, French DP, Bonetti D, Johnston DW. Assessment and measurement in health psychology. In: Sutton S, Baum A, Johnston M, editors. Handbook of health psychology. London: Sage; 2004. p. 288e323. [4] Johnston M. Mood in chronic disease: questioning the answers. Curr Psychol 1999;18:71e87. [5] Lister AM, Rode S, Farmer A, Salkovskis P. Does thinking about personal health risk increase anxiety? J Health Psychol 2002;7:409e14. [6] Farmer AJ, Doll HA. In a randomized trial, outcomes were not affected by intensive follow-up over 1 year. J Clini Epidemiol 2005;58:991e6.
[7] Farmer AJ, Doll H, Levy JC, Salkovskis PM. The impact of screening for Type 2 diabetes in siblings of patients with established diabetes. Diabetic Med 2003;20:996e1004. [8] Waugh N, Scotland G, McNamee P, Gillett M, Brennan A, Goyder E, et al. Screening for type 2 diabetes: literature review and economic modelling. Health Technol Assess 2007;11:17. [9] Adriaanse MC, Snoek FJ. The psychological impact of screening for type 2 diabetes. Diabetes Metab Res Rev 2006;22:20e5. [10] Eborall HC, Griffin SJ, Prevost AT, Kinmonth AL, French DP, Sutton S. Psychological impact of screening for type 2 diabetes: controlled trial and comparative study embedded in the ADDITION (Cambridge) randomised controlled trial. BMJ 2007;335(7619): 486e9. [11] Griffin SJ, Little PS, Kinmonth AL, Hales CN, Wareham NJ. Diabetes risk score: towards earlier detection of type 2 diabetes in general practice. Diabetes Metab Res Rev 2000;16:164e71. [12] Idler EL, Benyamini Y. Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behav 2007;38: 21e37. [13] Marteau TM, Bekker H. The development of a six-item short-form of the state scale of the Spielberger state-trait anxiety inventory (STAI). Br J Clin Psychol 1992;31:301e6. [14] Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand 1983;67:361e70. [15] Lerman C, Trock B, Rimer BK, Jepson C, Brody D, Boyce A. Psychological side effects of breast cancer screening. Health Psychol 1991;10:259e67. [16] Weinstein ND. Perceptions of personal susceptibility to harm. In: Mays VM, Albee GW, Schneider SF, editors. Primary prevention of AIDS: Psychological approaches. Newbury Park, CA: Sage; 1989. p. 142e67. [17] Weinstein ND, Diefenbach MA. Percentage and verbal category measures of risk likelihood. Health Educ Res 1997;12:139e41. [18] Moss-Morris R, Weinman J, Petrie KJ, Horne R, Cameron LD, Buick D. The revised illness perception questionnaire (IPQ-R). Psychol Health 2002;17:1e16. [19] Timman R, Roos R, Maat-Kievit A, Tibben A. Adverse effects of predictive testing for Huntington disease underestimated: long-term effects 7-10 years after the test. Health Psychol 2004;23:189e97. [20] Maissi E, Marteau TM, Hankins M, Moss S, Legood R, Gray A. The psychological impact of human papillomavirus testing in women with borderline or mildly dyskaryotic cervical smear test results: 6-month follow up. Br J Cancer 2005;92:990e4.