Research Methods. Paddy Gillespie a, *, Eamon O Shea a, Susan M Smith b, Margaret E Cupples c and Andrew W Murphy d. Abstract

Transcription

1 Family Practice, 2016, Vol. 33, No. 6, doi: /fampra/cmw088 Advance Access publication 1 September 2016 Research Methods A comparison of medical records and patient questionnaires as sources for the estimation of costs within research studies and the implications for economic evaluation Paddy Gillespie a, *, Eamon O Shea a, Susan M Smith b, Margaret E Cupples c and Andrew W Murphy d a School of Business and Economics, National University of Ireland, Galway, Ireland, b Department of General Practice, Royal College of Surgeons, Dublin, Ireland, c Department of General Practice and Primary Care, UKCRC Centre of Excellence for Public Health Research, Queen s University Belfast, Belfast, Northern Ireland, UK and d School of Medicine, National University of Ireland, Galway, Ireland. *Correspondence to Paddy Gillespie, School of Business and Economics, J. E. Cairnes Building, National University of Ireland (NUI), Galway, Ireland; paddy.gillespie@nuigalway.ie Abstract Background. Data on health care utilization may be collected using a variety of mechanisms within research studies, each of which may have implications for cost and cost effectiveness. Objective. The aim of this observational study is to compare data collected from medical records searches and self-report questionnaires for the cost analysis of a cardiac secondary prevention intervention. Methods. Secondary data analysis of the Secondary Prevention of Heart Disease in General Practice (SPHERE) randomized controlled trial (RCT). Resource use data for a range of health care services were collected by research nurse searches of medical records and self-report questionnaires and costs of care estimated for each data collection mechanism. A series of statistical analyses were conducted to compare the mean costs for medical records data versus questionnaire data and to conduct incremental analyses for the intervention and control arms in the trial. Results. Data were available to estimate costs for 95% of patients in the intervention and 96% of patients in the control using the medical records data compared to 65% and 66%, respectively, using the questionnaire data. The incremental analysis revealed a statistically significant difference in mean cost of 796 (95% CI: 1447, 144; P-value: 0.017) for the intervention relative to the control. This compared to no significant difference in mean cost (95% CI: 1446, 860; P-value: 0.619) for the questionnaire analysis. Conclusions. Our findings illustrate the importance of the choice of health care utilization data collection mechanism for the conduct of economic evaluation alongside randomized trials in primary care. This choice will have implications for the costing methodology employed and potentially, for the cost and cost effectiveness outcomes generated. Key words: Atherosclerosis, cardiovascular disorders, health economics, primary care. The Author Published by Oxford University Press. All rights reserved. For permissions, please journals.permissions@oup.com. 733

2 734 Family Practice, 2016, Vol. 33, No. 6 Introduction Issues of efficiency and cost effectiveness are becoming increasingly relevant for health policy decision making in primary care. The technique of economic evaluation is concerned with the provision of evidence to inform the efficient allocation of health care resources. Central to conduct of this technique is the process by which resource use data are identified, measured and valued for the purposes of analysing the costs of alternative health care programmes under consideration. For economic evaluation conducted alongside randomized controlled trials (RCT), the data collection process for resource utilization may be conducted prospectively or retrospectively and using mechanisms such as medical record chart searches, case report forms, self-report questionnaires, interviews or diaries (1). Each mechanism differs in its application and may result in varying degrees of data completeness and bias. In a systematic review of 42 studies that evaluated the accuracy of self-reported utilization data, Bhandari and Wagner (2) identified a series of factors that affect data accuracy including sample population and cognitive abilities, recall time frame, type of utilization, utilisation frequency, questionnaire design, mode of data collection and memory aids and probes. It follows, therefore, that the choice of data collection mechanism has important implications for the results from economic evaluation and, potentially, health policy decision making. In practice, although contemporaneous self-reporting is typically an effective and timeefficient data collection mode, there is evidence that older people tend to under report their utilization (3,4). Furthermore, reporting errors among this cohort have been shown to be particularly associated with longer recall periods and when the frequency of events increases (5). Such concerns give rise to the need for careful consideration of choice of data collection mechanism for older cohorts, particularly if the data are used to support economic evaluation. In this study, we explicitly consider this question in the context of a RCT which examined the cost effectiveness of a secondary prevention intervention for people with heart disease in primary care on the island of Ireland (6). In particular, data collected from medical record chart searches and self-report questionnaires from the SPHERE (Secondary Prevention of Heart Disease in General Practice) study is analysed to examine the implications of both approaches for the cost analysis and, by extension, the economic evaluation to consider the cost effectiveness of the SPHERE intervention. Methods The SPHERE randomized controlled trial We report full details on the RCT methods elsewhere (6). In summary, we conducted a cluster RCT of 48 practices and 903 patients with coronary heart disease [documented myocardial infarction (MI), coronary artery bypass grafting, angioplasty or angina] in both health care systems on the island of Ireland. Ethical approval was provided by the local committees at the participating study centres. Informed consent was obtained from each participant. Geographically, 16 practices were recruited from Belfast in Northern Ireland and 16 from both Dublin and Galway in the Republic of Ireland. Notably, people in Northern Ireland have free access, via the National Health Service (NHS), to GP care and prescriptions are issued free of charge, or with minimal cost. In the Republic of Ireland, only the one-third of the population with the lowest income and those in specified at-risk groups are covered by the publicly funded General Medical Services (GMS) scheme and entitled to free primary care, including prescriptions. Non-GMS patients must pay fully for visits to a GP and up to a monthly maximum limit for prescriptions. In the RCT, practices, and their patients, were randomized to the intervention group or to the control group. In the intervention group, practices and patients had access to the SPHERE intervention, consisting of tailored care plans in which practices received training in prescribing and behaviour change, administrative support and a quarterly newsletter, while patients received motivational interviewing, goal identification and target setting for lifestyle change and four monthly review visits. In the control group, patients received existing secondary prevention services in primary care, consisting of unstructured and irregular patient follow-up (6). From a clinical effectiveness perspective, the intervention was associated with a significant reduction in the numbers admitted to hospital (1.56; 95% CI: ) over the trial follow-up period of 1.5 years (7). From a cost effectiveness perspective, the intervention was, on average, less costly by 513 (95% CI: 1087, 92) and more effective by (95% CI: , ) quality-adjusted lifeyears (QALYs) at 1.5 years (8,9). Taken together, the results suggest that the SPHERE intervention is likely to be considered as a clinically and cost effective alternative to existing primary care for patients with coronary heart disease. Cost analysis The economic evaluation was conducted following the guidelines for health technology assessment for Ireland (10). With respect to the cost analysis, the perspective of the health care provider was adopted. Costs relating to the use of primary and secondary health care services over the course of the trial were estimated for patients in both treatment arms. This included the costs of GP and practice nurse (PN) consultations, outpatient (OPT), hospital admissions (INPT) and accident and emergency (AE) visits for any reason. A total cost variable was constructed by aggregating across all the individual resource costs. For the published economic evaluation, resource use data were captured by research nurses from computerized practice surgery records at baseline for 12 months prior to allocation, and for 18-month follow-up time period. A vector of unit costs was applied to calculate the cost associated with each resource activity reported for each method. Unit cost data for the relevant resource use items were collected for both Northern Ireland and the Republic of Ireland and were assigned on the basis of the patient s country of origin (Table 1). In sensitivity analysis, a series of alternative unit cost estimates were employed. Data collection For the purposes of this analysis, we focus on 18-month follow-up period of the RCT to compare resource use cost data collected by research nurses from medical records and resource use cost data generated by returned self-report questionnaires from the participants. At 18-month follow-up, resource use data was captured for all study participants by trained SPHERE research nurses from computerized practice surgery records (after obtaining patient consent) and recorded onto SPHERE study database (6). In addition, questionnaires and pre-paid envelopes were sent to all participating patients at 18-months follow-up from practices with administrative assistance from the SPHERE research nurse. Notably, the questionnaire was developed specifically for the patient cohort in SPHERE study. The research nurses liaised with practice staff to follow-up nonresponders. Six weeks was allowed for non-response, during which time reminders were posted. Of the 903 participants in the SPHERE study, 832 or 92.1% of participants returned self-report questionnaires at 18-month (11).

3 A comparison of medical records and patient questionnaires 735 Statistical analysis The statistical analysis comprised of a series of univariate analytical approaches to examine the valid responses for each data collection mechanism for each of the individual resource cost and total cost variables for study participants, adopting approaches which were informed by the hierarchical nature of the SPHERE dataset. First, the baseline characteristics for the patients who made up the medical records data and questionnaire data cohorts were compared using independent t-tests for continuous variables and chi-square tests for categorical variables. Second, descriptive statistics, in the form of means and standard deviations were estimated for all the cost variables of interest. Third, we estimate a series of paired t-tests to compare the estimated mean costs for those participants who have both medical records data and questionnaire data. Fourth, we conducted incremental analysis to estimate the difference in mean cost between the intervention and control arms for each of resource cost and total cost variables for both the medical records data and the questionnaire data. The incremental analysis was conducted on an intention to treat basis applying multilevel statistical techniques which recognize the clustering in the cost data (12,13). The results from the analyses described above were used to identify statistically significant associations for the response variables of interest, with statistical significance set at P <0.05 for all analyses. The statistical analyses were conducted using the statistics package Stata 13. Results The baseline characteristics for those in the full patient cohort for whom medical record data were collected and for the patient cohort who returned the self-report questionnaires are presented in Table 2. Table 1. Resource categories and unit costs for healthcare services for CHD secondary prevention in Republic of Ireland and Northern Ireland Resource item North Ire Source Rep of Ire Source Intervention set up 213 Study accounts 213 Study accounts Health care resources 46 Netten and Curtis (2006) 45 Revenue Commissioner Office GP visit 11 Netten and Curtis (2006) 7 Irish Nurses Organisation PN visit 186 DHSSPS 171 DOHC OPT visit 709 DHSSPS 659 DOHC INPT day 154 DHSSPS 210 DOHC AE visit AE visit, accident and emergency clinic; CHD, coronary heart disease; INPT, inpatient admission; North Ire, Northern Ireland; OPT, outpatient clinic; PN, practice nurse; Rep of Ire, Republic of Ireland. Source References: Netten A, Curtis J. (2006). Unit costs of health and social care. Canterbury: Personal Social Services Research Unit, University of Kent; Office of Revenue Commissioner, Dublin; Irish Nurses Organisation, Dublin; DHSSPS, Department of Health, Social Services and Public Safety; DOHC, Department of Health and Children. All prices reported in 2006 Euros ( ). Table 2. Summary statistics by data collection mechanism of Coronary Heart Disease patient characteristics for clinical risk factors, cardiovascular treatments, health systems and socioeconomic variables at baseline in the SPHERE study Variable (A) Medical chart cohort, N = 903 (B) Questionnaire only cohort, N = 832 (A) versus (B) Mean (SD)/% Mean (SD)/% P-value Clinical risk factor variables Age 68 (9) 67 (9) Gender: male Years since diagnosis 8 (7) 8 (7) History of MI History of diabetes Record of PTCA/CABG Systolic blood pressure > Diastolic blood pressure > Total cholesterol > Smoking status Body mass index > Cardiovascular treatment variables Intervention arm Lipid-lowering therapy Antihypertensive therapy Anticoagulant therapy Health system variable Republic of Ireland Socioeconomic variables Marital status: married/cohabiting Education status: completed secondary Occupation status: manual CHD, coronary heart disease; MI, myocardial infection; PTCA, percutaneous transluminal coronary angioplasty; CABG, coronary artery bypass grafting. Statistical analysis based on independent t-tests for continuous variables and chi-square tests for categorical variables. Statistical significance identified at p < 0.05.

4 736 Family Practice, 2016, Vol. 33, No. 6 Those who did not return self-report questionnaires were more likely to be smokers, and single or divorced. Data on available responses for resource use and costs are presented in Table 3. Based on the medical records, data availability for intervention patients was 96% for GP visits, 96% for PN visits, 95% for OPT visits, 96% for INPT nights and 95% for AE visits, and for control patients was 97%, 97%, 96%, 97% and 97%, respectively. By comparison, questionnaire data availability for intervention patients were 74% for GP visits, 73% for PN visits, 74% for OPT visits, 74% for INPT nights and 75% for AE visits, and for control patients was 76%, 74%, 78%, 75% and 75%, respectively. With respect to total costs, data were available for 95% of intervention patients and 96% of control patients based on the medical records data compared to 65% of intervention patients and 66% of controls patient based on the questionnaire data. Descriptive statistics in Table 3 highlights the variations in results across the different data collection mechanisms. In particular, it reports the mean resource use and mean cost estimates for each resource item by treatment arm. It also reports the mean total cost estimates by treatment arm. With respect to total costs, the mean estimates were 2292 (SD: 4334) for the intervention arm and 3089 (SD: 5210) for the control arm based on the medical record data and 2625 (SD: 8429) for the intervention arm and 2918 (SD: 6603) for the control arm based on the questionnaire data. The results from the comparison of mean cost results for those with medical record data and questionnaire data are presented in Table 4. This was based on a complete case analysis to estimate, using paired t-tests, the difference in mean cost between medical record data and questionnaire data for each resource cost and for total cost. For the intervention arm, although respondent data yielded significantly lower costs of GP visits, PN visits and OPT visits in comparison to recorded data, their reported costs of INPT nights were significantly higher. For the control arm, respondent data yielded significantly lower costs of GP visits and OPT visits, but significantly higher costs of PN visits compared to those derived from the medical records. The results from the incremental cost analyses to compare the mean cost estimates for the intervention and control arms are presented in Table 5. Results are presented for each resource cost and for total cost for both the medical record data analysis and the questionnaire data analysis. The results from the medical records analysis indicate a statistically significant difference in mean cost between the intervention and control arms for PN visits, OPT visits, INPT nights and AE visits. Notably, the questionnaire data analysis only indicates a statistically significant difference in the mean cost of AE visits. Moreover, the magnitudes of the mean cost difference estimates were generally lower in the questionnaire analyses, apart from for AE visits. In terms of the total cost analysis, the mean cost for the intervention group was significantly lower than the control group by 796 (95% CI: 1447, 144; P-value: 0.017) based on the medical record analysis. This compared to no significant difference in mean cost in the questionnaire analysis. Finally, the results from a series of sensitivity analyses for the incremental cost analyses are presented in Supplementary Table 1. These go to confirm the results from the base-case analyses. Discussion Main findings Central to the conduct of economic evaluation of health care programmes delivered in primary care is the process by which resource use data is identified, measured and valued for the purposes of analysing costs of care. It follows that the data collection mechanism adopted may have important implications for health care resource allocation decision making in this context. This study compared data collected from the medical record chart searches and self-report patient questionnaires from the SPHERE study to examine the implications of adopting one approach or the other for the cost analysis of a primary care intervention for cardiac secondary prevention. The findings suggest that the choice of data collection mechanism has important implications for data completeness and availability, with medical records, as expected, superior to self-report questionnaires Table 3. Summary statistics of Coronary Heart Disease secondary prevention resource use and cost estimates over 18 months by treatment arm and by data collection mechanism in the SPHERE Study Variable/analysis Response, n (% of N) Resource use, mean (SD) Resource cost ( ), mean (SD) Total cost ( ), mean (SD) Medical chart data Intervention N = 444 Control N = 459 Intervention Control Intervention Control Intervention Control GP visits 426 (96) 444 (97) 8.3 (5.7) 7.6 (6.0) 374 (257) 344 (269) PN visits 426 (96) 444 (97) 4.6 (4.2) 1.8 (2.2) 40 (38) 16 (20) OPT visits 426 (96) 444 (97) 2.7 (3.6) 3.6 (4.1) 527 (683) 697 (804) INPT days 422 (96) 442 (97) 2.0 (6.2) 3.0 (7.4) 1299 (4032) 1945 (4749) AE visits 426 (96) 444 (97) 0.3 (0.6) 0.5 (0.9) 52 (111) 85 (167) Total cost 422 (95) 442 (96) 2292 (4334) 3089 (5210) Questionnaire data Intervention N = 444 Control N = 459 Intervention Control Intervention Control Intervention Control GP visits 328 (74) 349 (76) 5.8 (5.5) 5.9 (6.1) 263 (248) 267 (274) PN visits 323 (73) 339 (74) 4.2 (6.4) 3.2 (4.7) 36 (48) 27 (38) OPT visits 330 (74) 358 (78) 2.0 (2.9) 2.5 (3.9) 356 (505) 433 (675) INPT days 330 (74) 346 (75) 3.1 (12.0) 3.3 (9.1) 2067 (7965) 2220 (6024) AE visits 331 (75) 344 (75) 0.3 (0.8) 0.5 (1.1) 63 (158) 98 (205) Total cost 289 (65) 305 (66) 2625 (8429) 2918 (6603) AE, accident and emergency; CHD, coronary heart disease; INPT, hospital admissions; OPT, outpatient; PN, practice nurse; SD, standard deviation. Response data for the medical record data and questionnaire data; mean estimates for resource use and resource costs for medical record data and questionnaire data; and mean estimates for total costs for medical record data and questionnaire data. All prices reported in 2006 Euros ( ).

5 A comparison of medical records and patient questionnaires 737 Table 4. Comparative Coronary Heart Disease secondary prevention cost analysis over 18 months by data collection mechanism in the SPHERE Study Variable/analysis Mean cost (SD) ( ) Mean cost difference ( ) Medical chart (M) minus questionnaire (Q) (95% CIs) (P-value) Resource item Total Intervention Control Total Intervention Control n = 675 n = 328 n = 347 GP visits M 350 (260) 368 (256) 332 (263) GP visits Q 265 (262) 263 (248) 267 (275) (63, 106) (0.000) (76, 136) (0.000) (33, 96) (0.000) n = 660 n = 323 n = 337 PN visits M 29 (24) 42 (40) 17 (21) PN visits Q 32 (44) 36 (48) 28 (38) ( 6, 1) (0.191) (1, 12) (0.013) ( 15, 7) (0.000) n = 686 n = 330 n = 356 OPT visits M 615 (759) 529 (702) 694 (801) OPT visits Q 397 (601) 356 (505) 435 (676) (160, 276) (0.000) (99, 248) (0.000) (171, 348) (0.000) n = 672 n = 329 n = 343 INPT days M 1452 (4318) 967 (3533) 1919 (4916) INPT days Q 2113 (6977) 1985 (7837) 2236 (6048) ( 1108, 213) (0.004) ( 1803, 234) (0.011) ( 768, 134) (0.167) n = 673 n = 331 n = 342 AE visits M 67 (143) 51 (112) 82 (166) AE visits Q 81 (184) 63 (158) 99 (206) ( 27, 2) (0.028) ( 27, 2) (0.096) ( 38, 5) (0.124) n = 591 n = 289 n = 302 Total cost M 2371 (4674) 1881 (3921) 2840 (5259) Total cost Q 2785 (7559) 2625 (8429) 2939 (6632) ( 931, 102) (0.115) ( 1655, 167) (0.109) ( 613, 415) (0.706) AE, accident and emergency; CHD, coronary heart disease; INPT, hospital admissions; OPT, outpatient; PN, practice nurse; SD, standard deviation. The results from a complete case analysis for the differences in mean costs, estimated using pair t-tests, for each resource cost and for total cost between medical record data and questionnaire data All prices reported in 2006 Euros ( ). Table 5. Incremental Coronary Heart Disease secondary prevention cost analysis over 18 months by data collection mechanism Variable/analysis Medical chart data Questionnaire data Resource item n Mean cost difference n Mean cost difference Intervention minus control Intervention minus control (95% CIs) (P-value) (95% CIs) (P-value) GP visits ( 32, 89) (0.354) ( 45, 38) (0.864) PN visits (14, 35) (0.000) ( 3, 19) (0.135) OPT visits ( 307, 17) (0.029) ( 179, 21) (0.120) INPT days ( 1245, 44) (0.035) ( 1125, 818) (0.757) AE visits ( 55, 10) (0.005) ( 63, 8) (0.012) Total cost ( 1447, 144) (0.017) ( 1446, 860) (0.619) AE, accident and emergency; CHD, coronary heart disease; INPT, hospital admissions; OPT, outpatient; PN, practice nurse. The results from the incremental cost analysis for the intervention versus the control, estimated using multilevel multivariate regression, for each resource cost and for total cost based on medical record data and questionnaire data. in terms of the generation of data. We also find significant differences in mean cost estimates depending on the data collection mechanism used. Furthermore, we find the data collection mechanism had important implications for the incremental cost analyses. In this case, a statistically significant mean difference in total cost was found using medical record data analysis, but not in the questionnaire data analysis. Our findings illustrate the importance of the choice of health care utilisation data collection mechanism for the conduct of economic evaluation alongside randomized trials in primary care. This choice will have implications for the costing methodology to be employed and potentially for the cost and cost effectiveness outcomes generated.

6 738 Family Practice, 2016, Vol. 33, No. 6 Other literature Our article adds to the literature that compares the accuracy of selfreported with administrative data on health care utilisation among older persons by exploring this issue for a cohort of patients with coronary heart disease and present results for the cost for a range of health care services and across treatment arms within a RCT. Notably as reflected across the literature, our findings contradict some studies but are consistent with others with respect to the under and over reporting of cost estimates. More specifically, Wallihan et al. (3) found that older adults substantially under report health service usage of hospitalisation and emergency room visits. Wolinsky et al. (4) reported that concordance between self-reported and claims-based hospital episodes was high, but concordance for physician visits was low. Van Dalen et al. (5) reported that self-reports of hospitalizations and GP visits in a broadly healthy community-dwelling older population seemed adequate and efficient. However, they also found that as people became older and more functionally impaired, self-report data were more prone to measurement bias. Finally, Hoogendoorn et al. (14) compared self-report versus care provider registrations for cost and cost utility analysis based on RCT. They found that while adoption of the alternative methods affected the within-treatment group cost comparisons, they did not affect the between treatment group comparisons of cost-of-cost utility. In our study, we find implications for both within and between treatment group comparisons. Limitations A number of points should be noted as potential limiting factors of this study. The analysis of questionnaire data focused on a period of recall of 18 months and it is likely that a shorter time horizon would go to eliminate some of the bias reported here (3). Furthermore, the questionnaire was designed specifically for the patient cohort in the SPHERE study and, as a result, was not a validated measurement tool. This may cast some doubt over the implications of our results. With respect to generalisability, persons with heart disease who were institutionalized, who typically have higher levels of utilisation (5), were excluded from the RCT. Furthermore, although the cost analysis was conducted from the health service perspective, certain resource items were not captured. For example, the costs of prescriptions, private patient costs and broader costs to society such as productivity losses were not captured in the analysis. In addition, the process of conducting cost analysis in Ireland is compromised by the lack of nationally available unit cost data. In estimating unit costs for individual resource activities, we endeavoured at all times to be conservative in any assumptions adopted. It should also be noted that we adopt 2006 prices for the analysis and medical inflation has fluctuated over the period since the trial was conducted. We employed appropriate methods for the univariate statistical analysis of cost data collected alongside cluster RCTs. However, we do not account here for covariate imbalances between treatment arms (13). Furthermore, missing data are an important consideration in any analysis and, in practice; multiple imputation methods may be undertaken (15). That said, our primary motivation with this article was to present the variation in responses and costs from the two data collection methods for illustrative purposes. Finally, it is important to note that data collection based on the medical records searches as is not without its own limitations (5). Conclusions Our results have important implications for the design of data collection mechanisms for health care resource utilisation alongside RCTs for purposes of conducting economic evaluation in primary care. Although not conclusive given the limited context, they do suggest that the choice of data collection mechanism may impact upon economic evaluation estimates and resource allocation decision making at the margin. In particular, while conducting medical record searches may not always be practical due to time, travel and financial constraints, facilitating such a process of data collection may increase the likelihood of obtaining valid and robust statistical results. Therefore, where possible, access to medical records should be made easier for researchers and records need to be more complete and better integrated to allow optimal use of the data. Supplementary material Supplementary material is available at Family Practice online. Declaration Funding: Health Research Board (HRA-HSR ). The funders had no part in the design of the study; the collection, analysis and interpretation of the data; the writing of the report and the decision to submit the article for publication. Ethical approval: Irish College of General Practitioners and the Queen s University Research Ethics Committee. Conflict of interest: none. Acknowledgements We would like to thank the patients and practitioners who participated in the study and to acknowledge the research nurses in the Republic of Ireland and the Northern Ireland Clinical Research Network (Primary Care) who supported the data collection. Trial registration: Current Controlled Trials ISRCTN PG, EOS, MEC, SMS, and AWM contributed to study conception and design. PG and EOS were responsible for the acquisition of data in the SPHERE Study while PG analysed the data. PG and EOS drafted the article and all authors revised the article and granted final approval to the version submitted for publication. References 1. Drummond MF, Sculpher MJ, Torrance GW, O Brien J, Stoddart GL. Methods for the Economic Evaluation of Health Care Programmes. London: Oxford University Press, Bhandari A, Wagner T. Self-reported utilization of health care services: improving measurement and accuracy. Med Care Res Rev 2006; 63: Wallihan DB, Stump TE, Callahan CM. Accuracy of self-reported health services use and patterns of care among urban older adults. Med Care 1999; 37: Wolinsky FD, Miller TR, An H et al. Hospital episodes and physician visits: the concordance between self-reports and medicare claims. Med Care 2007; 45: van Dalen M, Suijker J, MacNeil-Vroomen J et al. Self-report of healthcare utilization among community-dwelling older persons: a prospective cohort study. PLoS One 2014; 9: e Murphy AW, Cupples ME, Smith SM et al. The SPHERE Study. Secondary Prevention of Heart Disease in General Practice: protocol of a randomised controlled trial of tailored practice and patient care plans with parallel qualitative, economic and policy analyses [ISRCTN ]. Curr Control Trials Cardiovasc Med 2005; 6: Murphy AW, Cupples ME, Smith S et al. Secondary Prevention of Heart Disease in General Practice: a cluster randomised controlled trial of tailored practice and patient care plans. BMJ 2009; 339: b4220.

7 A comparison of medical records and patient questionnaires Gillespie P, O Shea E, Murphy AW et al. The cost effectiveness of the SPHERE intervention for the secondary prevention of coronary heart disease. Int J Technol Assess Health Care 2010; 26: Gillespie P, O Shea E, Murphy AW et al. Relative cost effectiveness of the SPHERE intervention in selected patient subgroups with existing coronary heart disease. Eur J Health Econ 2011; 13: Health Information and Quality Authority (HIQA). Guidelines for the Economic Evaluation of Health Technologies in Ireland, (accessed on 1 August 2015). 11. Cole JA, Gillespie P, Smith SM et al. Using postal questionnaires to evaluate physical activity and diet behaviour change: case study exploring implications of valid responder characteristics in interpreting intervention outcomes. BMC Res Notes 2014; 7: Campbell MK, Elbourne DR, Altman DG. CONSORT statement: extension to cluster randomised trials. BMJ 2004: 328: Gomes M, Ng ES, Grieve R et al. Developing appropriate methods for cost-effectiveness analysis of cluster randomized trials. Med Decis Making 2012; 32: Hoogendoorn M, van Wetering CR, Schols AM, Rutten-van Molken MP. Self-report versus care provider registration of healthcare utilization: impact on cost and cost-utility. Int J Technol Assess Health Care 2009; 25: Gomes M, Díaz-Ordaz K, Grieve R, Kenward M. Multiple imputation methods for handling missing data in cost-effectiveness analyses that use data from hierarchical studies an application to cluster randomized trials. Med Decis Making 2013; 33: