Family Practice Advance Access originally published online on February 18, 2005
Family Practice 2005 22(2):205-214; doi:10.1093/fampra/cmi009
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The ‘number needed to sample’ in primary care research. Comparison of two primary care sampling frames for chronic back pain
a Department of General Practice and Primary Care, University of Aberdeen, b Department of Public Health, University of Aberdeen, c Aberdeen Royal Infimary
Correspondence to Dr Blair H Smith, Department of General Practice and Primary Care, University of Aberdeen, Foresterhill Health Centre, Westburn Road, Aberdeen AB25 2AY, UK; Email: blairsmith{at}abdn.ac.uk
Received 20 August 2004; Accepted 16 September 2004.
Smith BH, Hannaford PC, Elliott AM, Smith WC and Chambers WA. The ‘number needed to sample’ in primary care research. Comparison of two primary care sampling frames for chronic back pain. Family Practice 2005; 22: 205–214.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Background. Sampling for primary care research must strike a balance between efficiency and external validity. For most conditions, even a large population sample will yield a small number of cases, yet other sampling techniques risk problems with extrapolation of findings.
Objective. To compare the efficiency and external validity of two sampling methods for both an intervention study and epidemiological research in primary care—a convenience sample and a general population sample—comparing the response and follow-up rates, the demographic and clinical characteristics of each sample, and calculating the ‘number needed to sample’ (NNS) for a hypothetical randomized controlled trial.
Methods. In 1996, we selected two random samples of adults from 29 general practices in Grampian, for an epidemiological study of chronic pain. One sample of 4175 was identified by an electronic questionnaire that listed patients receiving regular analgesic prescriptions—the ‘repeat prescription sample’. The other sample of 5036 was identified from all patients on practice lists—the ‘general population sample’. Questionnaires, including demographic, pain and general health measures, were sent to all. A similar follow-up questionnaire was sent in 2000 to all those agreeing to participate in further research. We identified a potential group of subjects for a hypothetical trial in primary care based on a recently published trial (those aged 25–64, with severe chronic back pain, willing to participate in further research).
Results. The repeat prescription sample produced better response rates than the general sample overall (86% compared with 82%, P < 0.001), from both genders and from the oldest and youngest age groups. The NNS using convenience sampling was 10 for each member of the final potential trial sample, compared with 55 using general population sampling. There were important differences between the samples in age, marital and employment status, social class and educational level. However, among the potential trial sample, there were no demographic differences. Those from the repeat prescription sample had poorer indices than the general population sample in all pain and health measures.
Conclusions. The repeat prescription sampling method was approximately five times more efficient than the general population method. However demographic and clinical differences in the repeat prescription sample might hamper extrapolation of findings to the general population, particularly in an epidemiological study, and demonstrate that simple comparison with age and gender of the target population is insufficient.
Keywords. Chronic back pain, clinical trials, postal surveys, primary care, sampling methods.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Sampling for intervention and epidemiological studies in primary care represents a challenge for researchers. The balance has to be struck between efficiency of sampling and the ability to generalise findings. There are probably four main methods of recruitment in primary care (Fig. 1).
|
Recruiting from the general population, by surveying a representative sample to identify cases of the condition. For example, the Bristol Helicobacter Project approached 27536 adults through their general practices, and eventually identified 1634 positive H pylori cases for a randomized controlled trial (RCT) and epidemiological study.1 This method minimizes the risk of selection bias, and maximizes the ability to apply findings to the general population with the disease or condition. However, it is inefficient, requiring the survey of a large number of individuals to generate a relatively small sample. As well as significant practical and resource implications, there are potential ethical difficulties with involving many people in research which is unlikely to be directly relevant to them.
Recruiting by invitation from a targeted convenience sample, such as those on a practice disease register. For example, a study of an asthma intervention could recruit from those in one or a few practices listed as receiving regular inhaled beta-agonists.2 This method has a higher risk of selection bias, but will generally improve the case identification rate. A relative lack of knowledge about the denominator hampers prevalence calculations from this kind of sampling, and the ability to generalize findings.
Recruiting patients with the condition under investigation opportunistically as they attend for treatment. For example, an intervention in lateral epicondylitis recruited from consecutive attenders with this condition in a group of practices in England.3 Recruitment rates can be relatively low, and the process difficult. It generally requires several enthusiastic practitioners to be actively involved in recruitment, takes a long time, risks selection bias,4 and suffers from an unknown denominator. In the epicondylitis study3 it took 37 GPs two years to recruit 164 subjects. Although the authors estimated that the effect of selection bias was minimal, they were unable to identify eligible patients not recruited, and therefore were unable to estimate incidence rates or confirm representativeness of their findings.
Sampling patients from those referred to secondary care. For example, a study of home based care for people with chronic obstructive pulmonary disease recruited from patients who had attended or been admitted to hospital with the disease.5 For most conditions treated in primary care, those who attend hospital clinics represent a small, selected proportion of all patients in the practice with the condition under investigation, as demonstrated at pain clinics.6 Findings from an intervention study which used this sampling frame may therefore be difficult to apply to all patients attending primary care, where conditions will generally be less severe, but where management of other co-existing conditions must also be considered.7
Other methods of sampling in primary care include voluntary methods (such as response to advertisements,8 recruiting from the waiting room,9 or use of electronic databases.10 These may be useful for some purposes, but have a high risk of selection bias.
The target population, that is the group of people to whom the findings are to be generalized, will determine the sample requirements. For example, the sample for an intervention study such as a RCT should ideally be representative of people with the condition for which the intervention is intended, or at least any important differences should be known. On the other hand, a sample for an epidemiological study should be representative of the population at risk, or any important differences should be known, if information on the distribution and determinants of the condition under study are to be extrapolated meaningfully. Targeted sampling represents a compromise between the need for representativeness and the need for efficient and ethical sampling. However, the relative efficiency and the effects of this compromise are unknown.
We compared the characteristics of samples recruited by two methods in primary care for research on chronic pain, one a targeted convenience sample and the other a general population sample, aiming to examine the utility of each for both an intervention study and for epidemiological research. We compared the efficiency of each method at identifying individuals for a hypothetical RCT, by calculating the ‘number needed to sample’ (NNS) in order to arrive at the required sample size. We also compared the response rates and follow-up rates four years later, and demographic and health-related characteristics of those identified by each sampling method, as indicators of their respective efficiency and utility for cross-sectional and cohort studies. This analysis is intended to inform consideration of sampling methods for future studies in primary care, and the extrapolation of results from existing studies to the general primary care population.
Research on chronic pain in primary care has used most of the sampling methods described above.11–14 A recent RCT of acupuncture in back pain recruited consecutive in-patients aged 20–60 with low back pain at a rehabilitation centre,11 with a calculated required sample size of 380. The findings of the study cannot be extrapolated directly to a primary care population, but the intervention could be. For our hypothetical study, we therefore tested the ability of our samples to identify individuals from the general primary care population aged 25–64 with severe chronic back pain who stated that they were willing to participate in further research. (The slight difference in age ranges was necessary as our research involved adults aged 25 years and above. Although such restriction by age may impose a further limitation on the ability to extrapolate to the majority of individuals with chronic back pain,4 our intention was for this hypothetical study to reflect a recent example.) How would our different methods of recruitment affect the conduct and results of a primary care based study of acupuncture in adults with severe chronic back pain?
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
In 1996 we initiated a cohort study of chronic pain, designed to examine its prevalence, distribution, onset and recovery in the community. We selected two samples of patients from 29 general practices in Grampian, Scotland. All practices in Grampian who used the General Practice Administration System for Scotland (GPASS) computer system were invited to participate (n = 67), and 29 agreed to do so. The GPASS system is widely used throughout Scotland, and one of its main uses is the administration of repeat prescriptions.15
The 29 practices were sent an electronic questionnaire (EQ) computer disk,15 which interrogated the practice GPASS database, and downloaded names and addresses of all patients recorded as being in receipt of regular prescriptions for an analgesic [all drugs listed in the British National Formulary16 Section 4.7 (analgesics), Section 10.1.1 (non-steroidal anti-inflammatory drugs) and Section 10.3.2 (rubefacients and other topical anti-rheumatics) except legally controlled drugs such as morphine (prescriptions for which cannot be generated legally by computer in the UK)]. Previous work17 had shown that only a very small proportion of patients who receive regular pain-related medication would be missed by not searching for controlled drugs, nor for aspirin, tri-cyclic anti-depressants or anti-convulsants, all of which are primarily prescribed regularly for non-painful conditions. This approach provided a targeted ‘pain-rich’ sampling frame, from which 4175 individuals aged over 25 years were randomly sampled, stratified by age, gender and registered practice. We have called this the ‘repeat prescription’ sample.
The Community Health Index (CHI) is a list of all patients registered with any GP in Scotland, probably around 98% of the population.18 Each patient is identified by a unique 10-digit number, and recorded details including the name, address, gender, date of birth and registered GP of each individual. The CHI relating to the populations registered with the 29 participating practices was obtained. A further 5036 individuals aged over 25 years were sampled from this, similarly stratified by age, gender and registered practice. Everyone in each practice aged over 25, including those receiving regular analgesic prescriptions, were eligible to be selected for this sample. Anyone already identified in the repeat prescription sample was replaced by the next name on the CHI list, to avoid duplicate mailing of questionnaires. The sample drawn from the CHI was intended to represent the general population, and we have called this the ‘general population’ sample.
Both sample lists were reviewed by the GPs, in order to remove individuals who were considered unsuitable for mailing (for example because of terminal or severe psychiatric illness). GPs were not asked to specify a reason for any exclusions. A questionnaire was sent to everyone remaining on the lists, with a covering letter signed by a member of the research team and a GP from the practice. Up to two reminders were sent to non-responders. The questionnaire included questions to identify the presence of chronic pain (‘persistent or intermittent pain or discomfort present for three months or longer’19,20); the reported cause of any chronic pain (back pain or sciatica, arthritis, injury, women's problems, angina, cancer, other, or unknown); the Chronic Pain Grade (CPG) questionnaire;21 the SF-36 general health questionnaire;22,23 and socio-demographic indicators (marital status, housing tenure, occupation, employment status, level of formal education). Respondents' age and gender were already identified during the sampling process. Social class was determined according to the Registrar General's occupational classification24 where this was possible from the information provided. This was only possible for those in employment at the time of survey, and provided classifications from social class 1 (higher professional occupations) to 5 (unskilled manual occupations). Housing tenure was used as a substitute for social class, with council-rented accommodation representing the lowest category.25 The CPG classifies chronic pain severity into four grades according to its perceived intensity and pain-related disability: Grade I (low intensity-low disability); Grade II (low intensity-high disability); Grade III (high intensity-high disability-moderately limiting); and Grade IV (high intensity-high disability-severely limiting). It was designed and validated in a primary care setting in USA21 with its validity for postal research confirmed in the UK.26 Each of the eight SF-36 dimensions has a maximum score of 100, this representing good health. Finally, all respondents were asked whether they would be willing to participate in further research.
A follow-up survey was conducted approximately four years later, in 2000. A list of all respondents who had agreed to participate in further research was reviewed by the GPs. Patients were excluded if they had left the practice (and their address was therefore unknown), if they had died, or for other unspecified reasons as before. The questionnaire in the follow-up survey was similar to that used in the 1996 survey, but included the reported site (rather than cause) of any chronic pain (back; neck or shoulder; head, face or teeth; stomach or abdomen; limbs; chest; or other).27
The 1996 and 2000 surveys were both approved by the Grampian Research Ethics Committee. Full details of the questionnaires, and epidemiological findings from the general population sample have been reported previously.28–30 Respondents with chronic back pain were identified as those responding positively to chronic pain identification questions, and identifying back pain as the cause of chronic pain in 1996, or the back as the site of chronic pain in 2000. Demographic characteristics and response and follow-up rates of the two samples were compared using chi-square testing. Scores on the SF-36 and CPG were compared using the Mann–Whitney U test for non-parametric data.
Corrected response rates were calculated as the number of questionnaires returned, divided by (the number posted minus the number returned by the postal service as sent to an incorrect address minus the number returned with a statement that the addressee was unable to complete it because of, for example, death or incapacity).
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Efficiency of sampling
The proportions excluded and responding in 1996 differed between the samples, with the repeat prescription sample being generally more efficient at each stage (Table 1). A total of 3335 individuals responded from the repeat prescription sample and 3605 from the general population sample, representing corrected response rates of 86.5% and 82.3%, respectively. In the repeat prescription sample 2019 respondents (60.5% of respondents) agreed to participate in further research, compared with 2422 (67.2%) in the general population sample. After 4 years, 1610 from the repeat prescription sample could be re-surveyed (48.3% of original respondents), of whom 1387 responded to the follow-up questionnaire (corrected response rate 86.1%). In the general population sample, a total of 1937 could be re-surveyed (53.7% of original respondents), of whom 1608 responded to the follow-up questionnaire (corrected response rate 83.0%).
|
There was a total of 248 people in the repeat prescription sample aged between 25 and 64 who had had severe (CPG III or IV) chronic back pain in 1996 and who were willing to participate in further research, compared with 60 in the general population sample (Table 2). To achieve a sample size of 380 people aged between 25 and 64 with severe chronic back pain for our hypothetical intervention study, the original repeat prescription sample size would need to have been 3780 (380/248 x 2467) for the repeat prescription sample (an NNS of 9.9 for each eligible subject). This compared with 20786 (380/60 x 3282) for the general population sample (an NNS of 54.7 for each eligible subject).
|
Corrected response rates in 1996 and rates of follow-up in 2000, among those aged 25–64, were compared between the two recruitment methods by age and sex (Table 3). The repeat prescription sample had higher response rates overall, from both gender groups, and from the oldest and youngest age-groups, and was therefore more representative of its population. The rate of follow-up, however, was the same in both samples (50.0%).
|
External validity. There were highly significant differences in the demographic characteristics of the repeat prescription sample (aged 25–64) compared with the general population sample (Table 4). The repeat prescription sample had lower proportions of respondents from younger age-groups, those still living as married, higher socio-economic groups, those in employment and with higher educational levels than the general population sample. However, respondents identified from the repeat prescription sample, aged 25–64 with severe chronic back pain who agreed to participate in further research were demographically similar to those identified from the general population sample, with the only significant difference found to be in employment status (Table 4).
|
The associated general health was compared between respondents from the repeat prescription and general population samples (aged 25–64) (Table 5). There was poorer health among the repeat prescription sample in all dimensions measured, whether all respondents were considered, or only those aged 25–64 with severe chronic back pain who agreed to participate in further research. This was the case both in the 1996 and in the 2000 survey. Among those with chronic pain, this was found on the CPG to be more severe in the repeat prescription sample than in the general population sample (P < 0.001, Mann–Whitney U testing). This was the case in both 1996 and 2000 for all respondents, and in 2000 only for those aged 25–64 with severe chronic back pain.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
The challenges associated with recruitment for large-scale surveys and RCTs are well-recognized31–34 although little research has been specific to primary care. This study has compared the effects of two different methods of sampling patients for primary care research, both for an intervention study and for epidemiological research. Although it was not designed specifically for this purpose, the study provides an indication of potential case identification, response and follow-up rates for primary care research. Both methods used strategies that are associated with increased response rates,34 and produced reasonable results, but the targeted convenience (repeat prescription) sample was approximately five times more efficient at identifying potential subjects for our hypothetical intervention study.
The ‘number needed to sample’ (NNS) could be defined as the number of people sampled for an initial inquiry to produce each subject that participates in the final or main study (which may be an RCT, but may also be more intensive data collection or interview). This takes account of response rates, eligibility according to the main study criteria, and agreement to participate in the main study. The NNS can be calculated as the initial sample size divided by the final sample size, and provides an indication of the efficiency of the initial sampling method.
For our hypothetical RCT, which, although not intended as a model study, was based on a recent example,11 the NNS was 9.9 for the repeat prescription sample, compared with 54.7 for the general population sample. These figures assume that everyone in our samples who agreed to participate in further research would consent to participate in the hypothetical RCT, and are probably artificially low as a consequence.33 Both of our sampling methods were able to identify individuals' ages before sampling. Other general population sampling frames (such as the electoral roll), or targeted convenience sampling frames (such as prescriptions received by a pharmacist) may not allow the identification of age. If the age of those in the sample were not known, the NNS for the repeat prescription sample would have been 16.8, and in the general population sample it would have been 83.9. If calculations were based on retention in the trial after four years, the NNS would be 18.0 and 121.6, respectively, if ages were known prior to sampling (and 30.4 and 186.5, respectively, if ages were not known). There are therefore likely to be considerable economic benefits in targeted convenience sampling for primary care research, and in using primary care sampling frames. There are, of course, ethical and resource implications inherent in trying to recruit patients for an RCT irrespective of whether they have attended a GP with the symptoms or condition under question, and these need to be considered before embarking on such an approach. In this respect, targeted convenience sampling will not seek the participation of so many individuals for whom the research is not directly relevant, or who are unlikely to benefit from it, and may be preferable. Our NNS figures relate to chronic back pain, which is a relatively common condition in the community.28 The NNS would be much higher for rarer conditions, and the gulf in NNS between targeted convenience and general population sampling greater still. Many RCTs in primary care require multi-centre sampling techniques with cluster randomisation,35 with greater sample size requirements36 and further inflation of the NNS.
Another complication arises from changes to the data protection legislation and guidance since we sampled for our study in 1996.37,39 Contacting individuals identified through the CHI and similar databases now requires the intercession of a ‘Caldicott guardian’, independent of the research team, who makes initial contact with potential subjects and only provides their personal details to the researchers after signed consent to do so is obtained. Although experience in response rates since this introduction is limited, initial indications are pessimistic, with response rates from 25%–60% obtained in Grampian; consent rates have been higher (around 60%–80%) for targeted samples (V Angus, L Iversen, pers. comms). This seems likely to further inflate NNS, particularly from general population sampling frames, and strengthens the case relatively for targeted convenience sampling.
Considering the relevance for epidemiological research, there were higher response rates overall and in the oldest and youngest age groups from the repeat prescription sample than the general population sample, and the repeat prescription sample was therefore more representative of its population. These differences may indicate a greater willingness for individuals to participate in research that seems to be relevant to them. However, the follow-up rates were the same in both samples, with the relatively good health found in the general population sample at follow-up possibly making their responses more likely. This information will inform sampling for cohort studies.
Simple comparison of the age and gender of a study population with that of the general population, in attempt to examine representativeness, may be insufficient and misleading, and subsequent extrapolation inappropriate without caveat. We found important demographic differences between each sample. It is likely that our general population sample represented the populations of the participating practices (after stratification), and broad comparison with the Grampian population supports this.39 The extent of these differences, and the number of variables affected suggest that a study population selected by targeted convenience sampling may be very different from the general population, and this would need to be acknowledged in epidemiological research.
There were, however, no demographic differences found between the two sampling methods among those aged 25–64 with severe chronic back pain who agreed to participate in further research (the hypothetical RCT sample), with the exception of employment status (Fig. 2). This may reflect similarities related to the presence of similar pathology, and supports the external validity of targeted convenience sampling for an RCT. The health of those in the repeat prescription sample was poorer, though, and the pain tended to be more severe than those in the general population sample, even among those in the hypothetical study sample. These differences would be important in extrapolating the results of an RCT to a general primary care population, which will include older people and others not fulfilling the inclusion criteria for this hypothetical study, yet for whom an intervention may be considered. Depending on the intervention, poorer initial health, perhaps with co-morbidities, may render an intervention aimed specifically at chronic back pain (for example physical therapy) less likely to be beneficial, and underestimate its effectiveness in the general population. On the other hand, because of a greater overall potential for benefit with poorer initial health, an intervention aimed at a broader view of health (such as cognitive behavioural therapy or alternative therapy) may exaggerate the effectiveness in a population with less severe disease.
|
Conclusion
We have compared two methods of sampling in primary care, based on repeat prescribing and general population sampling frames, using a hypothetical intervention study in chronic back pain. We have demonstrated that there are considerable advantages of efficiency in using targeted convenience sampling, which is associated with a lower ‘number needed to sample’. However, there are likely to be some important health and demographic differences that will limit generalisation of results pertaining to an intervention tested in a targeted convenience sample to a general primary care population. This paper illustrates what some of these differences might be. Extrapolations based only on age and gender may be misleading. In all studies, the sampling technique and consequent interpretation of results must be fit for the purpose.
|
Declaration |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Funding: this research was funded by the Scottish Office Home and Health Department, Chief Scientist Office, and by the Association of Anaesthetists of Great Britain and Northern Ireland. BS is supported by an NHS R&D Primary Care Career Scientist Award, funded by the Scottish Executive Health Department, Chief Scientist Office.
Ethical approval: the 1996 and 2000 surveys were both approved by the Grampian Research Ethics Committee.
Conflicts of interest:
|
Acknowledgments |
|---|
We are grateful to all the GPs who assisted with the sampling process, to Hilary Selbie and Marisa Haetzman who assisted with data collection in 2000, and to Netta Clark who conducted data entry.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
1 Murray L, Johnston B, Lane A, Harvey I, Donovan J, Nair P, Harvey R. Relationship between body mass and gastro-oesophageal reflux symptoms: the Bristol Helicobacter Project. Int J Epidemiol 2003; 32: 645–650.
2 Pinnock H, Bawden R, Proctor S, Wolfe S, Scullion J, Price D, Sheikh A. Accessibility, acceptability, and effectiveness in primary care of routine telephone review of asthma: pragmatic, randomised controlled trial. Br Med J 2003; 326: 477–479.
3 Hay EM, Paterson SM, Lewis M, Hosie G, Croft P. Pragmatic randomised controlled trial of local corticosteroid injection and naproxen for treatment of lateral epicondylitis of elbow in primary care. Br Med J 1999; 319: 964–968.
4 Mant D. Can randomised trials inform clinical decisions about individual patients? Lancet 1999; 353: 743–746.[CrossRef][Web of Science][Medline]
5 Hermiz O, Comino E, Marks G, Daffurn K, Wison S, Harris M. Randomised controlled trial of home based care of patients with chronic obstructive pulmonary disease. Br Med J 2002; 325: 938–942.
6 Crombie IK, Davies HTO. Selection bias in pain research. Pain 1998; 74: 1–3.[CrossRef][Web of Science][Medline]
7 Rosser WW. Application of evidence from randomised controlled trials to general practice. Lancet 1999; 353: 661–664.[CrossRef][Web of Science][Medline]
8 Gerard SF, Smith BH, Simpson JA. A randomised controlled trial of spiritual healing in restricted neck movement. Journal of Alternative and Complementary Medicine 2003; 9: 467–478.[CrossRef][Web of Science][Medline]
9 Richardson J, Coid J, Petruckevitch A, Chung WS, Moorey S, Feder G. Identifying domestic violence: cross sectional study in primary care. Br Med J 2002; 324: 274–280.
10 Price D, Thomas M, Mitchell G, Niziol C, Featherstone R. Improvement of asthma control with a breath-actuated pressurised metred dose inhaler (BAI): a prescribing claims study of 5556 patients using a traditional pressurised metred dose inhaler (MDI) or a breath-actuated device. Respir Med 2003; 97: 12–19.[CrossRef][Web of Science][Medline]
11 Molsberger AF, Mau J, Pawalec DB, Winkler J. Does acupuncture improve the orthopedic management of chronic low back pain—a randomized, blinded, controlled trial with 3 months follow up. Pain 2002; 99: 579–587.[CrossRef][Web of Science][Medline]
12 Serpell MG, Neuropathic Pain Group. Gabapentin in neuropathic pain syndromes: a randomised, double-blind, placebo-controlled trial. Pain 2002; 99: 557–566.[CrossRef][Web of Science][Medline]
13 Moore JE, Von Korff M, Cherkin D, Saunders K, Lorig K. A randomized trial of a cognitive-behavioral program for enhancing back pain self care in a primary care setting. Pain 2000; 88: 145–153.[CrossRef][Web of Science][Medline]
14 Gureje O, Von Korff M, Simon GE, Gater R. Persistent pain and well-being. A World Health Organization study in primary care. J Am Med Assoc 1998; 280: 147–151.
15 Taylor MW, Ritchie LD, Taylor RJ, Ryan MP, Paterson NIA, Duncan R, Brotherston KG. General practice computing in Scotland. Br Med J 1990; 300: 170–172.
16 British National Formulary, Number 32. London: British Medical Association and Royal Pharmaceutical Society of Great Britain; 1995.
17 Smith BH, Read JRM, Chambers WAC, Watt B, Grimshaw JM. Researching chronic pain: identification of a community based sample. The Pain Clinic 1996; 9: 73–76.
18 Bowling A. Research Methods in Health: Investigating Health and Health Services. Buckingham: Open University Press; 1997.
19 International Association for the Study of Pain. Classification of chronic pain. Pain 1986; Suppl 3: S1–S225.[CrossRef]
20 Purves AM, Penny KI, Munro C, Smith BH, Grimshaw J, Wilson B, Smith WC, Chambers WA. Defining chronic pain for epidemiological research—assessing a subjective definition. The Pain Clinic 1998; 10: 139–147.
21 Von Korff M, Ormel J, Keefe FJ et al. Grading the severity of chronic pain. Pain 1992; 50: 133–149.[CrossRef][Web of Science][Medline]
22 Ware JE, Snow KK, Kosinski M et al. SF 36 Health Survey. Manual and Interpretation Guide. Boston, MA: The Health Institute; 1993.
23 Garratt AM, Ruta DA, Abdalla MI, Buckingham JK et al. The SF 36 health survey questionnaire: an outcome measure suitable for routine use within the NHS? Br Med J 1993; 306: 1440–1444.
24 Office of Population Censuses and Surveys. Standard Occupational Classification. London: HMSO; 1991.
25 Stoate HG. Can health screening damage your health? J Royal Coll Gen Pract 1989; 39: 189–195.
26 Smith BH, Penny K, Purves A, Munro C, Smith WC, Grimshaw J, Wilson B, Chambers WAC. The Chronic Pain Grade questionnaire: acceptability and validity in postal research. Pain 1997; 71: 141–147.[CrossRef][Web of Science][Medline]
27 Von Korff M. Epidemiologic and survey methods: chronic pain assessment. In Turk DC, Melzack R (eds). Handbook of Pain Assessment. 2nd Edition New York: Guildford Press, 2001; pp. 603–618.
28 Elliott AM, Smith BH, Penny KI Chambers WA, Smith WC. The epidemiology of chronic pain in the community. Lancet 1999; 354: 1248–1252.[CrossRef][Web of Science][Medline]
29 Smith BH, Elliott AM, Chambers WA, Smith WC, Hannaford PC, Penny K. The impact of chronic pain in the community. Fam Pract 2001; 18: 292–299.
30 Elliott AM, Smith BH, Hannaford PC, Smith WC, Chambers WA. The course of chronic pain in the community: results of a four-year follow-up study. Pain 2002; 99: 299–307.[CrossRef][Web of Science][Medline]
31 Moser CA, Kalton G. Survey Methods in Social Investigation, 2nd edn. Aldershot: Dartmouth; 1971.
32 Hunninghake DB, Darby CA, Probstfield JL. Recruitment experiences in clinical trials: literature summary and annotated bibliography. Controlled Clinical Trials 1987; 8(suppl 4): S6–S30.
33 Prescott RJ, Counsell CE, Gillespie WJ, Grant AM, Russell IT, Kiauka S, Colthart IR, Ross S, Shepard SM, Russell D. Factors that limit the quality, number and progress of randomized controlled trials. Health Technology Assessment 1999; 3: 20.
34 Edwards P, Roberts I, Clarke M, DiGuiseppe C, Pratap S, Wentz R, Kwan I. Increasing response rates to postal questionnaires: systematic review. Br Med J 2002; 324: 1183–1191.
35 Underwood M, Hannaford P, Slowther A. Masterclasses in Primary Care Research. Number 4, Randomised Controlled Trials and Multi-Centre Research. London: Royal College of General Practitioners; 2000.
36 Underwood M, Barnett A, Hajioff S. Cluster randomization: a trap for the unwary. Br J Gen Pract 1998; 48: 1089–1090.[Web of Science][Medline]
37 The Caldicott Committee. Report on the review of patient-identifiable information—December 1997. London: NHS Executive; 1999.
38 Medical Research Council. MRC Ethics Series. Good Research Practice. London: MRC; 2000.
39 General Register Office for Scotland. Scotland's Census, April 29, 2001. Available at http://www.gro-scotland.gov.uk/grosweb/grosweb.nsf/pages/file5/$file/key_stats_chbareas.pdf (accessed 6 May, 2003).
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  N. Torrance, B. H Smith, M. C Watson, and M. I Bennett Medication and treatment use in primary care patients with chronic pain of predominantly neuropathic origin Fam. Pract., October 1, 2007; 24(5): 481 - 485. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||