- Email: [email protected]
Use of a cross-sectional survey to estimate outcome of health care
Journal of Clinical Epidemiology 54 (2001) 1112–1119
Use of a cross-sectional survey to estimate outcome of health care: The example of anxiety and depression Umesh T. Kadam*, Peter Croft, Martyn Lewis Keele University, Primary Care Sciences Research Centre, Staffordshire, UK Received 7 August 2000; received in revised form 24 April 2001; accepted 26 April 2001
Abstract Our study proposes that a population-based cross-sectional survey can be used to estimate the outcome of health care by linking general practice morbidity records to the survey. Using the example of anxiety and depression to test this idea, we conducted a survey of an adult population registered with one general practice in the UK. The Hospital Anxiety and Depression (HAD) questionnaire was used to identify cases and controls. After mailing to a randomly selected adult population of 4002, there was an adjusted response rate of 66% (n 2,606), with 416 (16%) high-score cases, 506 (19%) medium-score cases, and 1684 (65%) low-score controls. All cases were compared with a sample of controls (n 450). In the 12 months before the survey, the high-score case group had experienced significantly higher GP contacts (n 377 [91%] versus 354 [79%]), diagnoses for anxiety or depression (119 [29%] versus 21 [5%]), and related drug treatments (111 [27%] versus 22 [5%]) compared with the control sample. Most of the diagnoses and drug treatments had been initiated at least 9 months before the survey. The linkage between the survey and the clinical records suggested that the health outcome of previously identified anxious and depressed patients was poor, with an estimated two-thirds who will not have fully recovered within an average of 9 months. This study demonstrates the potential for using cross-sectional population surveys to estimate not only the need for health care but also the outcome of health care. © 2001 Elsevier Science Inc. All rights reserved. Keywords: Cross-sectional studies; Outcome assessment; Epidemiologic methods; General practice; Anxiety and depression
1. Introduction Population surveys are an important epidemiological tool because they are based on the principle that the sample of the target population surveyed is unselected by health care setting or membership of particular groups. Such survey methods are used to estimate the health care needs for a given problem in a population. In the particular example of neurotic disorders, the Epidemiological Catchment Area study [1] and the UK OPCS survey [2] have estimated the prevalence in the population for these conditions at approximately 15% and this figure would represent the overall likely need for health care. Outcome of health care, in contrast, has been traditionally estimated by prospective studies, in particular randomized controlled trials. The generalizability of the results of trials is often limited by the selectivity of their participating populations, and a complementary method is to perform ob-
* Corresponding author. Keele University, Primary Care Sciences Research Centre, Staffordshire ST5 5BG, UK. Tel.: 1782-583-924; fax: 1782-583-911. E-mail address: [email protected] (U. Kadam)
servational follow-up studies to determine what happens to a broad group of patients who receive the treatment or intervention in actual practice. Examples of such approaches in the field of neurotic disorders include the World Health Organization (WHO) international study on depression outcome [3] and the study of the influence of adherence to antidepressant treatment guidelines on subsequent relapse and recurrence of depressive illness [4]. So traditionally, distinct and different approaches have been used to determine need and outcome. Assessing both are key objectives in delivering effective health care. Local health care systems have adopted the survey as one approach to measuring need, but methods to assess outcome of care are less easily translated from research to a health service environment. It is curious that epidemiological surveys are assumed to identify need for health care, but will actually include groups of people who have already received health care, and who at time of the survey have either recovered or not recovered, as well as people who have symptoms but have not received care. All of this means that the traditional survey of need is also presenting the observer with a picture of the outcome of health care already received in that population.
0895-4356/01/$ – see front matter © 2001 Elsevier Science Inc. All rights reserved. PII: S0895-4356(01)00 3 7 9 - 1
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
To use a survey both as a measure of outcome of health care as well as health need, we have to know the prior health care provided for the condition. In the UK, the clinical general practice records of the population are continuous, irrespective of whether the patients move practice, and are lifelong. More than 95% of the UK population are registered with general practitioners who provide access to most health care services within the National Health Service. Hence, these records can provide data on all clinical diagnoses and treatments in the registered population. By relating the previous health records of individuals to a cross-sectional survey, the survey results will represent an estimate of the outcome of earlier diagnoses and health care. Few studies have attempted to use retrospective case-control methods to assess outcome of health care, except in the field of screening [5] and isolated investigations of other topics [6]. Our study used a postal survey to assess the prevalence of health care needs for anxiety and depression in a population registered with a general practice in the UK, and we hypothesized that such a survey could also be used to estimate the outcome of health care. We used general practice data as the source of historical information about health care, and applied it to the health status determined by the population survey, to estimate the current outcome of primary health care for anxiety and depression in this population.
1113
Practice morbidity database [7]. The software system for the morbidity recording and prescriptions was VAMP Vision [8]. Records of all practice contacts, both clinical and administrative, include information on who entered the data, where the consultation took place, and the reason and date of contact. 2.4. Phase 1: Population survey The survey was mailed to the selected sample, and two follow-ups were sent to nonresponders. There were a number of health-related questions within the survey, but the core questionnaire for this study was the Hospital Anxiety and Depression (HAD) scale [9]. The HAD scale detects problems in the preceding week and acts both as a screening tool for anxiety and depression and to chart progression of the disorder over time. The HAD scale has been validated and used in general medical settings, and is useful in the assessment of clinical group comparisons and in correlation studies with different diseases [10–12]. The questionnaire is scored separately for anxiety and depression, and the score can be categorized into low-score controls (score 0 to 7), medium-score cases (8 to 10), and high-score cases (11 or greater) [10,13]. The HAD scale was selected in preference to other primary care measures because it measures both anxiety and depression, which commonly occur together in primary care [14], and it is easy to self-complete.
2. Methods 2.1. Design The study was a population-based case-control study. There were two phases: phase 1 was a postal survey of the registered practice population and phase 2 a retrospective 12-month review of practice-held records of a sample of the surveyed population. Survey subjects who had current high or medium scores on a scale of anxiety or depression symptoms were compared with a control group of subjects with low scores. 2.2. Study population As more than 95% of the British population are registered with a general practitioner, the practice register forms an effective sampling frame of a local population, regardless of actual use of health care. All adults aged 18 years and over who were registered with one group general practice (n 8004) formed the sampling frame for the questionnaire survey. From this adult registered practice population, 4002 patients were randomly selected; i.e., 50% of the practice population formed the sample for survey mailing. 2.3. Study setting The study took place in a semi-urban general practice in North Staffordshire that had complete computerized records of all practice contacts since 1990. It is a recording practice for the Royal College of General Practitioners returns service, which provides data for the National UK General
2.5. Phase 2: Record review The record review covered a total period of 12 months before the survey, between the dates of 13.8.95 and 13.8.96. The study groups derived from the survey for the record review comprised all high-score cases, all medium-score cases, and a randomly selected sample of 450 controls from the low-score group. In addition, we investigated a random sample of 100 nonresponders to the survey to investigate possible nonresponse bias. Random selection was done using numbers generated by Epi-Info Statistical software [15]. The computer records of all the study groups were retrieved and all individual identifiers were removed to preserve confidentiality. Information downloaded included details on age, gender, and postal codes. The postal codes were used to determine deprivation status. The resulting deprivation score (the Townsend score) allocated to individuals was derived from the local small area status in the national UK census. The Townsend score uses data on housing quality, car ownership, and number of people in the household to produce a composite score of relative deprivation [16]. All patient contacts with the GP during the total time period of 12 months were identified. Any diagnoses given by the general practitioner for mental disorders were identified from the diagnostic coding of these consultations. These are based on the READ Code classification, a commonly used coding system within the National Health Service in the UK [17]. The specific Read E category (mental disorders) and
1114
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
Read 13 category (social and personal history) were selected as the most common categories relating to anxiety and depression. Excluding specific diagnoses distinctive from neurotic disorders, notably schizophrenia and drug abuse, further refined the diagnostic data. The record review also identified drug treatments related to anxiety and depression, and these were classified according to the British National Formulary (BNF) classifications. The specific categories were chapter 4.1 (hypnotics and anxiolytics) and chapter 4.3 (antidepressants) [18]. Drug treatment was the intervention that we assessed, as it is currently the most widely used therapy in the treatment of anxiety and depression in primary care [19,20]. Classification of the data obtained from records to represent “need for health care” was as follows: (i) GP contact for any problem was defined as an expression of need for health care by the patient; (ii) diagnosis of anxiety and depression was a professionally identified need for health care; and (iii) drug treatment was defined as a separate professional intervention to meet the need for health care. For the record analysis, GP contact was defined as at least one such contact for any reason during the study period; diagnosis of anxiety or depression was defined as at least one such recorded diagnosis; and drug treatment as at least one such recorded prescription in the study period. 2.6. Sample size and statistical analysis Approximately 7% of adults in the UK population consult with anxiety or depression annually [2]. To detect a doubling in previous anxiety or depression diagnostic rate in the high-score group, conditional on an estimated 7% diagnostic prevalence rate in the control group and 416 high-score cases, a sample size of 450 controls would be required when 0.05, power 90% and statistical testing is two-tailed. The proportions of survey-identified subjects who had experienced practice contacts during the previous 12 months are presented together with summary odds ratios to compare the two case groups separately with the control group for their association with prior GP contact, diagnosis, and drug treatment. Statistical analysis was performed using the chi-square test for trend to assess the association of HAD score with individual demographic characteristics. Multivariate analyses using unconditional logistic regression were used to compare data on GP contact, diagnoses, and drug treatments (independent variables) between each case group separately and the control group, after adjusting the odds ratios for confounding by age, gender, and Townsend deprivation score. Because the therapy may take up to 3 months to have an effect, we separately performed subgroup analyses with respect to patients who had a record of anxiety or depression diagnosis or drug treatment at time intervals of at least 6 weeks, 3 months, and 6 months before the survey. Statistical significance is quoted for values of p 0.05 and all hypothesis testing was two-tailed. All analyses were per-
formed using SPSS version 9.0 for Windows [21]. The Local Research Ethics Committee, as part of the practice’s audit program, granted ethical approval to the practice for the survey and record review. 3. Results 3.1. Phase 1: Population survey In the total survey sample of 4002, there were 34 patients who were temporarily registered patients, i.e., less than 3 months with the practice, so the adjusted survey response rate after the exclusion of this group was 66% (n 2606). The number of patients with a high HAD score in the survey responders for anxiety or depression was 416 (16%); 506 (19%) patients had a medium score; and there were 1684 (65%) controls. The prevalence of probable anxiety and depression as shown by the high score was similar to that found by other surveys [10,11](see Table 1). 3.2. Phase 2: Retrospective record review 3.2.1. Case groups compared with control sample The total study group for record analysis comprised 1372 patients; all 416 high-score cases, all 506 medium-score cases, and a random sample of 450 (27%) controls. Demographically, the two case groups differed from the control subsample (Table 1). The case groups had a higher proportion of female patients but were similar in age. There was a trend of increasing Townsend deprivation scores (more deprivation) with increased anxiety or depression score. The number of patients who had had no contact with the GP were 39 (9%) of the high-score group, 74 (15%) of the medium-score group, and 96 (21%) of the control group. The case-control analysis of clinical records showed that there were significantly more GP contacts, anxiety or depression diagnoses, and related drug treatments for the highscore group compared with the control group (Table 2). In the 12-month study period, 32 patients (8%) in the highscore group had had a diagnosis without drug treatment and 24 (6%) had had a record of drug treatment without a diagnosis. Differences between the medium-score group and control group were less marked than for the high-score group versus control group comparison, but were still statistically significant (Table 2). In terms of an excluding diagnosis, the number of patients in each of the study groups were as follows: 6 (1%) in the high-score group, 11 (2%) in the medium-score group, and 5 (1%) in the control group. 3.2.2. Time interval between clinical record data and the survey In the 12-month study period, the time intervals between the first entry of diagnoses and drug treatments in the general practice records and the cross-sectional survey are shown in Table 3. The median interval between the first recorded anxiety or depression diagnosis, or the relevant drug treatment, and the date of the survey was 279 days for the
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
1115
Table 1 Demographic characteristics of the study groups Demographic characteristic Age
Gender Townsend score
Controls (n 450)
Medium-score (n 506)
High-score (n 416)
Category
n
%
n
%
n
%
18 to 29 30 to 44 45 to 59 60 to 75 Female Male 1.00 0.99 to 0.01 0 to 0.99 1.00
84 128 117 121 226 224 72 143 183 49
18.7 28.4 26.0 26.9 50.2 49.8 16.1 32.0 40.9 11.0
68 125 165 148 299 207 70 144 235 52
13.4 24.7 32.6 29.2 59.1 40.9 14.0 28.7 46.9 10.4
57 117 137 105 265 151 40 126 193 52
13.7 28.1 32.9 25.2 63.7 36.3 9.7 30.7 47.0 12.7
high-score group, 283 days for the medium-score group, and 288 days for the controls (Table 3). So the total study group had a median interval of at least 9 months between the first general practice record of anxiety or depression and the date of the survey. Similar findings emerged when diagnoses (median interval 266 days; quartile range 119 days to 329 days) and drug treatments (265 days; 109 to 340 days) were considered separately. Three subgroup case-control analyses were performed by excluding patients with a diagnosis or treatment record at least 6 weeks, 3 months, and 6 months before the survey, respectively. After adjusting for age, gender, and Townsend deprivation score, the differences between the case and control groups were still maintained. For the medium-score case group versus the control group, the diagnosis or treatment differences were as follows: 6 weeks (OR 2.19; 95% CI 1.40–3.42), 3 months (2.13; 1.34–3.41) and 6 months (2.06; 1.25–3.37). For the high-score case group versus the control group, the differences were as follows: 6 weeks (OR 6.24; 4.08–9.55), 3 months (6.31; 4.05–9.85), and 6-months (5.72; 3.57–9.17). 3.2.3. Estimating the population outcome The controls for the medical records analysis (n 450) represented a randomly selected subsample of patients with a low score (n 1684) in the survey, whereas the cases represented all those identified in the survey with a score of 8
P-value (2trend) 0.201
0.001 0.007
or greater. In this subsample control group, 33 had had a previous diagnosis or drug treatment for anxiety or depression (Table 4). So the projected number of controls with a previous record of anxiety or depression would be 123 in the whole survey low-score group (n 1684). Our definition of population outcome is as follows: subjects in the case groups with a prior record of anxiety and depression represent those who had not fully recovered, and subjects in the low-score group with a prior record represent those who had recovered. The numbers in the study that were identified with such records were: 143 in the highscore group, 82 in the medium-score group, and an estimated 123 in the noncase group. This means that, among all survey responders with a diagnosis or drug treatment for anxiety or depression in the previous year (projected total n 143 82 123 348), 65% (143 82 out of 348) had not fully recovered, and an estimated 35% (123 out of 348) had recovered. 3.2.4. Responders compared with nonresponders There were 1362 (34%) people who did not respond to the questionnaire survey. A randomly selected sample (n 100) from this group was evaluated separately to investigate nonresponse bias. There was little gender difference between the nonresponder and responder groups, but the nonresponders were younger and more likely to be from deprived areas. The record review of the 12-month period
Table 2 Retrospective comparison of GP contact, diagnosis, and drug treatment by study groups Clinical record GP contact Diagnosis or treatment Diagnosis Treatment a
c
No Yes Noc Yes Noc Yes Noc Yes
Controls
Medium score
High score
ORa (95% CI)
ORb (95% CI)
96 (21.3) 354 (78.7) 417 (92.7) 33 (7.3) 429 (95.3) 21 (4.7) 428 (95.1) 22 (4.9)
74 (14.6) 432 (85.4) 424 (83.8) 82 (16.2) 455 (89.9) 51 (10.1) 445 (87.9) 61 (12.1)
39 (9.4) 377 (90.6) 273 (65.6) 143 (34.4) 297 (71.4) 119 (28.6) 305 (73.3) 111 (26.7)
1.0 1.5 (1.1–2.1) 1.0 2.3 (1.5–3.5) 1.0 2.2 (1.3–3.7) 1.0 2.4 (1.4–4.0)
1.0 2.5 (1.7–3.7) 1.0 6.4 (4.3–9.7) 1.0 8.0 (4.9–13.1) 1.0 6.9 (4.2–11.2)
Odds ratio for medium-score group versus control group adjusted for age, gender, and Townsend deprivation score. Odds ratio for high-score group versus control group adjusted for age, gender, and Townsend deprivation score. c Reference category. b
1116
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
Table 3 Time interval (days) from first record of anxiety or depression to survey, by study groups Clinical record
Group
Sample (n)
Median (days to survey)
Quartiles (lower, upper)
Range (min, max)
Diagnosis or treatment
Controls Medium-score High-score Controls Medium-score High-score Controls Medium-score High-score
33 82 143 21 51 119 22 61 111
288 283 279 288 252 266 283 265 265
196, 345 159, 335 133, 337 179, 344 117, 330 119, 326 204, 349 135, 334 109, 341
39 to 365 7 to 364 1 to 365 39 to 365 13 to 364 1 to 365 39 to 365 7 to 364 1 to 365
Diagnosis
Treatment
before the survey for the nonresponder sample showed that 70% had had a GP contact and 12% had a record of diagnosis or drug treatment for anxiety or depression. The only significant difference in the record review for the nonresponders group compared with the responder group was that they had had less GP contact, but in terms of recorded history of diagnoses and treatment for anxiety or depression, there were no differences (Table 5). Applying the results of the record review to all 1362 nonresponders, the projected number of nonresponders with a previously recorded diagnosis or drug treatment for anxiety or depression is estimated to be 163. If we repeat the calculations of population outcome for the whole surveyed population, including nonresponders, the figures would be as follows: 511 in total with a relevant record; 123 (24%) patients who had recovered; 225 (44%) who had not fully recovered, and 163 (32%) patients whose recovery was not determined. 4. Discussion 4.1. Study findings: population survey as an estimate of health care need The questionnaire survey confirmed the high prevalence of anxiety and depression symptoms in the adult UK population. Figures for the high score group were comparable to those of other studies [1,2], which have estimated the prevalence of neurotic disorders such as anxiety and depression at approxi-
mately 15% in the population. The associated demographic characteristics also confirm results from other studies [1,2]. 4.2. Study findings: population survey as an estimate of health care outcome Patients with a high HAD score were more likely to have contacted the GP for any reason, at least once, in the prior 12 months than those with milder or no symptoms. This group also had the highest proportion of patients with previous diagnosis or drug treatment for anxiety and depression. Previously diagnosed patients had been identified a median of 9 months before the survey, and so clinical outcome as measured by the questionnaire survey did not represent recent onset. One interpretation of this result is that, regardless of any interventions between diagnosis and the subsequent survey, this group represents poor outcome despite diagnosis. Because most of these diagnosed patients had received a drug treatment, this group also represents poor outcome despite treatment. While the score of the high-score group indicates poor outcome for previously diagnosed or treated patients, the noncase score for previously identified patients indicates “recovered” cases. Projections from our study suggest that, of all those in the practice population who had been diagnosed or treated for anxiety or depression in the course of 12 months, an estimated 35% had fully recovered, compared with the 65% who had not improved. This means that one-
Table 4 Estimated population outcome for surveyed patients with a prior record (diagnosis or treatment) of anxiety or depression Survey responders Controls n 1684 Medium-score n 506 High-score n 416 Total n 2606 a
Sampled records
Prior record of anxiety or depression in sampled records
Estimated number with record in all survey responders (%; 95% CI)
Relative outcomea
450
33
123b (4.7; 3.2 to 6.3)
123/348 35%
506
82
82 (3.2; 2.5 to 3.8)
82/348 24%
416
143
143 (5.5; 4.8 to 6.2)
143/348 41%
1372
258
348
100%
Estimated number of patients in the three separate outcome groups (as defined by their survey score) expressed as a percentage of all patients with a prior record of anxiety or depression (n 348). b Estimated total number of controls with a prior record of anxiety or depression, calculated by applying the results of the sampled records i.e. 33/450 (7.3%) to the total of 1684.
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
1117
Table 5 Comparison of sociodemographic and medical record characteristics between responders and nonresponders to the survey
Age
Gender Townsend score
GP contact Diagnosis or treatment Diagnosis Treatment
18 to 29 30 to 44 45 to 59 60 to 75 Female Male 1.00 0.99 to 0.01 0.00 to 0.99 1.00 No Yes No Yes No Yes No Yes
Respondersa % (95% CI)
Nonrespondersb % (95% CI)
16.9 (14.4–19.3) 27.7 (24.8–30.5) 28.4 (25.6–31.2) 27.1 (24.2–29.9) 54.1 (50.9–57.3) 45.9 (42.7–49.1) 14.6 (12.3–16.9) 30.9 (27.9–33.9) 42.7 (39.6–45.8) 11.0 (9.0–13.0) 18.1 (15.6–20.7) 81.9 (79.3–84.4) 86.6 (84.8–88.5) 13.4 (11.5–15.2) 90.5 (88.9–92.0) 9.5 (8.0–11.1) 90.2 (88.7–91.8) 9.8 (8.2–11.3)
35.0 (25.7–44.3) 34.0 (24.7–43.3) 19.0 (11.3–26.7) 12.0 (5.6–18.4) 48.0 (38.2–57.8) 52.0 (42.2–61.8) 6.2 (1.4–11.0) 27.8 (18.9–36.8) 55.7 (45.8–65.6) 10.3 (4.3–16.4) 30.0 (21.0–39.0) 70.0 (61.0–79.0) 88.0 (81.6–94.4) 12.0 (5.6–18.4) 92.0 (86.7–97.3) 8.0 (2.7–13.3) 92.0 (86.7–97.3) 8.0 (2.7–13.3)
a Percentage figures are based on extrapolation of results derived from the total record review sample of 1372 to the 2606 responders in the survey (using weighted linear function). b Percentage figures are based on the 100 nonresponders randomly sampled from the total of 1362 nonresponders.
third of anxious or depressed patients in the community had had their needs for health care identified in primary care but had also had those needs met effectively. The two-thirds with poor outcome included those who had been treated with drugs in the previous twelve months and yet were still cases in the survey. This may indicate a need for better use of drug treatment [19] or for more suitable and effective interventions. The relative proportions of those who had improved or not only relates to patients who were consulters and excludes patients who had not contacted their GP. The implications for health care of this group of nonconsulters who had had no record of practice contact might concern, for example, issues such as access to care, rather than outcomes of care. There is evidence from other sources to support the finding that, despite identification in general practice of patients with anxiety and depression; these patients may not improve in the long term. In a study comparing depressed patients whose status was disclosed to GPs with patients whose condition was not disclosed, disclosure led to worse clinical outcomes [22]. In an international study on outcome, over half of those detected to be depressed were still depressed 12 months later [3]. In using effective interventions, one may need to consider that patients with anxiety or depression may have a chronic or inevitably relapsing and recurrent condition or a more severe illness. The findings for patients in a medium HAD score group are more open to interpretation. The clear difference was that a higher proportion of this group had had a GP contact compared with the control group. However, the proportion with an anxiety or depression diagnosis or related drug treatment was lower than that in the high-score group. Thus, the medium-score score for those who were previously diagnosed
and treated patients could represent either a partially “poor” outcome (compared with controls) or a partially “good” outcome (compared with the high-score group). 4.3. Study findings: limitations in interpretation The categorization of the HAD score into high and low scores as a screening instrument allows the differentiation between those who were likely to be “true” cases as measured by high score and those who were “true” noncases as measured by a low score. In terms of the study, this categorization was used to estimate the outcomes for previously diagnosed patients, but does not necessarily imply that survey-defined “caseness” and clinically defined “caseness” are the same [12]. The possibility of censorship limits the interpretation of the estimated population outcomes based on the case and control groups. First, patients who were lost to follow-up as a result of anxiety or depression could potentially contribute to a censorship bias. One example of such a bias, suicidal death, was considered, and no such deaths were identified in the study period. Second, there was a 34% nonresponse rate to the survey, and the demographic comparisons showed that responders and nonresponders were different. However, based on our projections (see Results section), even if all 163 nonresponders with a previous diagnosis or treatment of anxiety or depression had recovered by the time of the survey, at least 44% of previously diagnosed subjects will still not have fully recovered. 4.4. Cross-sectional survey as an estimate of outcome: critique and potential There are three general cautions if our approach is to be used more widely. First, the issue of censorship bias, which
1118
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
in the classic case-control study relates mainly to mortality. If the exposure under investigation causes death, the crosssectional study may underestimate any association between exposure and disease. For primary care epidemiology this may not often be an issue because many clinical outcomes relate to morbidity as opposed to mortality. However, nonresponders to the survey may also give rise to a similar bias. Second, cross-sectional surveys tend to pick up chronic cases, so that consultation and treatment for a 12-month period may reflect a longer period of ineffective or intermittent treatment, and so may be more about chronicity than about other acute or episodic problems. However, the control group provides a balancing estimate of beneficial clinical outcomes of both short- and longer-term problems. Third, the clinical outcome cannot be disentangled from potential unknown confounding factors. In our study, clinical outcome may have been related to unknown confounding factors other than treatment, although the results were adjusted for the important potential confounders of age, sex, and deprivation. This study was completed in a practice that had recorded all patient practice contacts and used standard diagnostic and drug treatment classification. Previous studies have validated the use of this type of computer data in general practice [23–25]. The clinical data were not standardized for the purposes of research but nevertheless captured the patient’s condition at time of contact as the general practitioner perceived it in actual practice. Therefore, all recorded clinical information has been accepted to possess “face validity.” The study highlights the advantages of well-organized, complete, and routinely collected general practice data. Cross-sectional population surveys are routinely used to estimate the need for health care, and yet have not been used for estimating outcome of health care by linking the survey to retrospective records. Our design paralleled the classic aetiological case-control study by being a cross-sectional study of current disease status, onto which we grafted an historical study of exposure. So the strengths of our method were its potential for practical application in local studies, and using a study population that was unselected and whose source of information was also unselected, because the general practitioner is the major gateway to all forms of health care in the UK. Our approach may provide a more realistic picture of the outcome of actual health care as it is provided and taken up within one general practice population, than one gleaned from the usual study of single interventions in selected populations. In conclusion, by combining the results of a cross-sectional survey with prior recorded clinical history, not only can health care needs be identified, but also the current effectiveness with which such needs may have been met can be estimated. There are caveats to the interpretation of results from this type of research method, and issues of selectivity of “chronic survivors” in the case group need to be addressed when considering generalizability. However, as a research method for primary care, the questionnaire survey
can thus not only be a static tool for estimating health care need but, by adding a time dimension from practice records, it can serve as a dynamic tool for providing a perspective on the outcome of such identified health needs. Acknowledgments North Staffordshire Health Authority supported UTK in a Public Health post during the project year and the Royal College of General Practitioners Scientific Foundation Board Grant funded part of the project. We are very grateful to all the patients and staff of the study practice. We would also like to thank Rob McCarney for his help in the survey data collection, Rhian Hughes for assistance in record data collection, and Paul Trinder for his comments on the earlier drafts of the paper. References [1] Robins LN, Regier DA. Psychiatric disorders in America. The Epidemiological Catchment Area Study. New York: Free Press, 1991. [2] Meltzer H, Gill B, Petticrew M, Hinks K. The prevalence of psychiatric morbidity among adults living in private households. OPCS Surveys of Psychiatric Morbidity in Great Britain: Report No.1. London: HMSO, 1995. [3] Goldberg D, Privett, Ustun B, Simon G, Linden M. The effects of detection and treatment on the outcome of major depression in primary care: a naturalistic study in 15 cities. Br J Gen Pract 1998;48:1840–4. [4] Melfi CA, Chawla AJ, Croghan TW, Hanna MP, Kennedy S, Sredl K. The effects of adherence to antidepressant treatment guidelines on relapse and recurrence of depression. Arch Gen Psychiatry 1998;55: 1128–32. [5] Clarke EA, Anderson TW. Does screening by “PAP” smears help prevent cervical cancer? A case-control study. Lancet 1979;2:1–4. [6] Jick H, Dinan N, Rothman KJ. Oral contraceptives and non-fatal myocardial infraction. JAMA 1978;239:1403–5. [7] Office of Population Census and Surveys (OPCS). Morbidity Statistics from General Practice. Fourth National Study 1991–1992. Series MB5 No. 3. London: HMSO. [8] VAMP Vision. Practice health systems limited. London; 1996. [9] Zigmond A, Snaith RP. The Hospital Anxiety and Depression Scale. Acta Psychiatr Scand 1983;67:361–70. [10] Wilkinson MJB, Barczak P. Psychiatric screening in general practice: comparisons of the General Health Questionnaire and the Hospital Anxiety and Depression Scale. J R Coll Gen Pract 1988;38:311–13. [11] Lewis G, Wessely S. Comparison of the General Health Questionnaire and the Hospital Anxiety and Depression Scale. Br J Psychiatry 1990;157:860–4. [12] Herrmann C. International experiences with the Hospital Anxiety and Depression scale—a review of validation data and clinical results. J Psychosom Res 1997;17–41. [13] Carroll BT, Kathol RG, Noyes R, Wald TG, Clamon GH. Screening for depression and anxiety in cancer patients, using the Hospital Anxiety and Depression Scale. Gen Hosp Psychiatry 1993;15:69–74. [14] Sartorius N, Ustun TB, Lecrubier Y, Wittchen H-U. Depression comorbid with anxiety: results from the WHO study on psychological disorders in primary health care. Br J Psych 1996;168(suppl 30):38–43. [15] Dean SG, Dean JA, Coulombier D, Brendel KA, Smith DC, Burton AH, Dicker RC, Sullivan K, Fagan RF, Arener TG. Epi-Info, Version 6: a word processing, database, and statistics program for epidemiology on microcomputers. Atlanta: Center for Disease Control and Prevention; 1994. [16] Townsend P, Phillimore P, Beattie A. Health and deprivation: Inequality and the North. London: Croom Helm, 1988.
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119 [17] NHS Centre for Coding and Classification. Read Codes and the Terms Projects: a Brief Guide. London: DoH, 1993. [18] British Medical Association and Royal Pharmaceutical Society of Great Britain. British National Formulary. London: British Medical Association/The Pharmaceutical Press, 1996. [19] Donoghue JM, Tylee A. The treatment of depression: prescribing patterns of antidepressants in primary care in the UK. Br J Psychiatry 1996;168:164–8. [20] Ohayon MM, Caulet M, Priest RG, Guilleminault C. Psychotropic medication consumption patterns in the UK general population. J Clin Epidemiol 1998;51:273–83. [21] Statistical Package for the Social Sciences (SPSS) Version 9.0. Chertsey (UK): SPSS Inc., 1998.
1119
[22] Dowrick C, Buchan I. Twelve month outcome of depression in general practice: does detection or disclosure make a difference? BMJ 1995;311:1274–6. [23] Wilson AE, Pollock C, Weekes T, Dowell A. Can general practices provide useful information? Evaluation of primary health care information project in Northern England. J Epidemiol Community Health 1995;49:227–30. [24] Jick H, Jick SS, Derby LE. Validation of information recorded on general practitioner based computerized data resource in the United Kingdom. BMJ 1991;302:766–8. [25] Pringle M, Ward P, Chilvers C. Assessment of the completeness and accuracy of computer medical records in four practices committed to recording data on computer. Br J Gen Pract 1995;45:537–41.
Use of a cross-sectional survey to estimate outcome of health care: The example of anxiety and depression Umesh T. Kadam*, Peter Croft, Martyn Lewis Keele University, Primary Care Sciences Research Centre, Staffordshire, UK Received 7 August 2000; received in revised form 24 April 2001; accepted 26 April 2001
Abstract Our study proposes that a population-based cross-sectional survey can be used to estimate the outcome of health care by linking general practice morbidity records to the survey. Using the example of anxiety and depression to test this idea, we conducted a survey of an adult population registered with one general practice in the UK. The Hospital Anxiety and Depression (HAD) questionnaire was used to identify cases and controls. After mailing to a randomly selected adult population of 4002, there was an adjusted response rate of 66% (n 2,606), with 416 (16%) high-score cases, 506 (19%) medium-score cases, and 1684 (65%) low-score controls. All cases were compared with a sample of controls (n 450). In the 12 months before the survey, the high-score case group had experienced significantly higher GP contacts (n 377 [91%] versus 354 [79%]), diagnoses for anxiety or depression (119 [29%] versus 21 [5%]), and related drug treatments (111 [27%] versus 22 [5%]) compared with the control sample. Most of the diagnoses and drug treatments had been initiated at least 9 months before the survey. The linkage between the survey and the clinical records suggested that the health outcome of previously identified anxious and depressed patients was poor, with an estimated two-thirds who will not have fully recovered within an average of 9 months. This study demonstrates the potential for using cross-sectional population surveys to estimate not only the need for health care but also the outcome of health care. © 2001 Elsevier Science Inc. All rights reserved. Keywords: Cross-sectional studies; Outcome assessment; Epidemiologic methods; General practice; Anxiety and depression
1. Introduction Population surveys are an important epidemiological tool because they are based on the principle that the sample of the target population surveyed is unselected by health care setting or membership of particular groups. Such survey methods are used to estimate the health care needs for a given problem in a population. In the particular example of neurotic disorders, the Epidemiological Catchment Area study [1] and the UK OPCS survey [2] have estimated the prevalence in the population for these conditions at approximately 15% and this figure would represent the overall likely need for health care. Outcome of health care, in contrast, has been traditionally estimated by prospective studies, in particular randomized controlled trials. The generalizability of the results of trials is often limited by the selectivity of their participating populations, and a complementary method is to perform ob-
* Corresponding author. Keele University, Primary Care Sciences Research Centre, Staffordshire ST5 5BG, UK. Tel.: 1782-583-924; fax: 1782-583-911. E-mail address: [email protected] (U. Kadam)
servational follow-up studies to determine what happens to a broad group of patients who receive the treatment or intervention in actual practice. Examples of such approaches in the field of neurotic disorders include the World Health Organization (WHO) international study on depression outcome [3] and the study of the influence of adherence to antidepressant treatment guidelines on subsequent relapse and recurrence of depressive illness [4]. So traditionally, distinct and different approaches have been used to determine need and outcome. Assessing both are key objectives in delivering effective health care. Local health care systems have adopted the survey as one approach to measuring need, but methods to assess outcome of care are less easily translated from research to a health service environment. It is curious that epidemiological surveys are assumed to identify need for health care, but will actually include groups of people who have already received health care, and who at time of the survey have either recovered or not recovered, as well as people who have symptoms but have not received care. All of this means that the traditional survey of need is also presenting the observer with a picture of the outcome of health care already received in that population.
0895-4356/01/$ – see front matter © 2001 Elsevier Science Inc. All rights reserved. PII: S0895-4356(01)00 3 7 9 - 1
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
To use a survey both as a measure of outcome of health care as well as health need, we have to know the prior health care provided for the condition. In the UK, the clinical general practice records of the population are continuous, irrespective of whether the patients move practice, and are lifelong. More than 95% of the UK population are registered with general practitioners who provide access to most health care services within the National Health Service. Hence, these records can provide data on all clinical diagnoses and treatments in the registered population. By relating the previous health records of individuals to a cross-sectional survey, the survey results will represent an estimate of the outcome of earlier diagnoses and health care. Few studies have attempted to use retrospective case-control methods to assess outcome of health care, except in the field of screening [5] and isolated investigations of other topics [6]. Our study used a postal survey to assess the prevalence of health care needs for anxiety and depression in a population registered with a general practice in the UK, and we hypothesized that such a survey could also be used to estimate the outcome of health care. We used general practice data as the source of historical information about health care, and applied it to the health status determined by the population survey, to estimate the current outcome of primary health care for anxiety and depression in this population.
1113
Practice morbidity database [7]. The software system for the morbidity recording and prescriptions was VAMP Vision [8]. Records of all practice contacts, both clinical and administrative, include information on who entered the data, where the consultation took place, and the reason and date of contact. 2.4. Phase 1: Population survey The survey was mailed to the selected sample, and two follow-ups were sent to nonresponders. There were a number of health-related questions within the survey, but the core questionnaire for this study was the Hospital Anxiety and Depression (HAD) scale [9]. The HAD scale detects problems in the preceding week and acts both as a screening tool for anxiety and depression and to chart progression of the disorder over time. The HAD scale has been validated and used in general medical settings, and is useful in the assessment of clinical group comparisons and in correlation studies with different diseases [10–12]. The questionnaire is scored separately for anxiety and depression, and the score can be categorized into low-score controls (score 0 to 7), medium-score cases (8 to 10), and high-score cases (11 or greater) [10,13]. The HAD scale was selected in preference to other primary care measures because it measures both anxiety and depression, which commonly occur together in primary care [14], and it is easy to self-complete.
2. Methods 2.1. Design The study was a population-based case-control study. There were two phases: phase 1 was a postal survey of the registered practice population and phase 2 a retrospective 12-month review of practice-held records of a sample of the surveyed population. Survey subjects who had current high or medium scores on a scale of anxiety or depression symptoms were compared with a control group of subjects with low scores. 2.2. Study population As more than 95% of the British population are registered with a general practitioner, the practice register forms an effective sampling frame of a local population, regardless of actual use of health care. All adults aged 18 years and over who were registered with one group general practice (n 8004) formed the sampling frame for the questionnaire survey. From this adult registered practice population, 4002 patients were randomly selected; i.e., 50% of the practice population formed the sample for survey mailing. 2.3. Study setting The study took place in a semi-urban general practice in North Staffordshire that had complete computerized records of all practice contacts since 1990. It is a recording practice for the Royal College of General Practitioners returns service, which provides data for the National UK General
2.5. Phase 2: Record review The record review covered a total period of 12 months before the survey, between the dates of 13.8.95 and 13.8.96. The study groups derived from the survey for the record review comprised all high-score cases, all medium-score cases, and a randomly selected sample of 450 controls from the low-score group. In addition, we investigated a random sample of 100 nonresponders to the survey to investigate possible nonresponse bias. Random selection was done using numbers generated by Epi-Info Statistical software [15]. The computer records of all the study groups were retrieved and all individual identifiers were removed to preserve confidentiality. Information downloaded included details on age, gender, and postal codes. The postal codes were used to determine deprivation status. The resulting deprivation score (the Townsend score) allocated to individuals was derived from the local small area status in the national UK census. The Townsend score uses data on housing quality, car ownership, and number of people in the household to produce a composite score of relative deprivation [16]. All patient contacts with the GP during the total time period of 12 months were identified. Any diagnoses given by the general practitioner for mental disorders were identified from the diagnostic coding of these consultations. These are based on the READ Code classification, a commonly used coding system within the National Health Service in the UK [17]. The specific Read E category (mental disorders) and
1114
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
Read 13 category (social and personal history) were selected as the most common categories relating to anxiety and depression. Excluding specific diagnoses distinctive from neurotic disorders, notably schizophrenia and drug abuse, further refined the diagnostic data. The record review also identified drug treatments related to anxiety and depression, and these were classified according to the British National Formulary (BNF) classifications. The specific categories were chapter 4.1 (hypnotics and anxiolytics) and chapter 4.3 (antidepressants) [18]. Drug treatment was the intervention that we assessed, as it is currently the most widely used therapy in the treatment of anxiety and depression in primary care [19,20]. Classification of the data obtained from records to represent “need for health care” was as follows: (i) GP contact for any problem was defined as an expression of need for health care by the patient; (ii) diagnosis of anxiety and depression was a professionally identified need for health care; and (iii) drug treatment was defined as a separate professional intervention to meet the need for health care. For the record analysis, GP contact was defined as at least one such contact for any reason during the study period; diagnosis of anxiety or depression was defined as at least one such recorded diagnosis; and drug treatment as at least one such recorded prescription in the study period. 2.6. Sample size and statistical analysis Approximately 7% of adults in the UK population consult with anxiety or depression annually [2]. To detect a doubling in previous anxiety or depression diagnostic rate in the high-score group, conditional on an estimated 7% diagnostic prevalence rate in the control group and 416 high-score cases, a sample size of 450 controls would be required when 0.05, power 90% and statistical testing is two-tailed. The proportions of survey-identified subjects who had experienced practice contacts during the previous 12 months are presented together with summary odds ratios to compare the two case groups separately with the control group for their association with prior GP contact, diagnosis, and drug treatment. Statistical analysis was performed using the chi-square test for trend to assess the association of HAD score with individual demographic characteristics. Multivariate analyses using unconditional logistic regression were used to compare data on GP contact, diagnoses, and drug treatments (independent variables) between each case group separately and the control group, after adjusting the odds ratios for confounding by age, gender, and Townsend deprivation score. Because the therapy may take up to 3 months to have an effect, we separately performed subgroup analyses with respect to patients who had a record of anxiety or depression diagnosis or drug treatment at time intervals of at least 6 weeks, 3 months, and 6 months before the survey. Statistical significance is quoted for values of p 0.05 and all hypothesis testing was two-tailed. All analyses were per-
formed using SPSS version 9.0 for Windows [21]. The Local Research Ethics Committee, as part of the practice’s audit program, granted ethical approval to the practice for the survey and record review. 3. Results 3.1. Phase 1: Population survey In the total survey sample of 4002, there were 34 patients who were temporarily registered patients, i.e., less than 3 months with the practice, so the adjusted survey response rate after the exclusion of this group was 66% (n 2606). The number of patients with a high HAD score in the survey responders for anxiety or depression was 416 (16%); 506 (19%) patients had a medium score; and there were 1684 (65%) controls. The prevalence of probable anxiety and depression as shown by the high score was similar to that found by other surveys [10,11](see Table 1). 3.2. Phase 2: Retrospective record review 3.2.1. Case groups compared with control sample The total study group for record analysis comprised 1372 patients; all 416 high-score cases, all 506 medium-score cases, and a random sample of 450 (27%) controls. Demographically, the two case groups differed from the control subsample (Table 1). The case groups had a higher proportion of female patients but were similar in age. There was a trend of increasing Townsend deprivation scores (more deprivation) with increased anxiety or depression score. The number of patients who had had no contact with the GP were 39 (9%) of the high-score group, 74 (15%) of the medium-score group, and 96 (21%) of the control group. The case-control analysis of clinical records showed that there were significantly more GP contacts, anxiety or depression diagnoses, and related drug treatments for the highscore group compared with the control group (Table 2). In the 12-month study period, 32 patients (8%) in the highscore group had had a diagnosis without drug treatment and 24 (6%) had had a record of drug treatment without a diagnosis. Differences between the medium-score group and control group were less marked than for the high-score group versus control group comparison, but were still statistically significant (Table 2). In terms of an excluding diagnosis, the number of patients in each of the study groups were as follows: 6 (1%) in the high-score group, 11 (2%) in the medium-score group, and 5 (1%) in the control group. 3.2.2. Time interval between clinical record data and the survey In the 12-month study period, the time intervals between the first entry of diagnoses and drug treatments in the general practice records and the cross-sectional survey are shown in Table 3. The median interval between the first recorded anxiety or depression diagnosis, or the relevant drug treatment, and the date of the survey was 279 days for the
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
1115
Table 1 Demographic characteristics of the study groups Demographic characteristic Age
Gender Townsend score
Controls (n 450)
Medium-score (n 506)
High-score (n 416)
Category
n
%
n
%
n
%
18 to 29 30 to 44 45 to 59 60 to 75 Female Male 1.00 0.99 to 0.01 0 to 0.99 1.00
84 128 117 121 226 224 72 143 183 49
18.7 28.4 26.0 26.9 50.2 49.8 16.1 32.0 40.9 11.0
68 125 165 148 299 207 70 144 235 52
13.4 24.7 32.6 29.2 59.1 40.9 14.0 28.7 46.9 10.4
57 117 137 105 265 151 40 126 193 52
13.7 28.1 32.9 25.2 63.7 36.3 9.7 30.7 47.0 12.7
high-score group, 283 days for the medium-score group, and 288 days for the controls (Table 3). So the total study group had a median interval of at least 9 months between the first general practice record of anxiety or depression and the date of the survey. Similar findings emerged when diagnoses (median interval 266 days; quartile range 119 days to 329 days) and drug treatments (265 days; 109 to 340 days) were considered separately. Three subgroup case-control analyses were performed by excluding patients with a diagnosis or treatment record at least 6 weeks, 3 months, and 6 months before the survey, respectively. After adjusting for age, gender, and Townsend deprivation score, the differences between the case and control groups were still maintained. For the medium-score case group versus the control group, the diagnosis or treatment differences were as follows: 6 weeks (OR 2.19; 95% CI 1.40–3.42), 3 months (2.13; 1.34–3.41) and 6 months (2.06; 1.25–3.37). For the high-score case group versus the control group, the differences were as follows: 6 weeks (OR 6.24; 4.08–9.55), 3 months (6.31; 4.05–9.85), and 6-months (5.72; 3.57–9.17). 3.2.3. Estimating the population outcome The controls for the medical records analysis (n 450) represented a randomly selected subsample of patients with a low score (n 1684) in the survey, whereas the cases represented all those identified in the survey with a score of 8
P-value (2trend) 0.201
0.001 0.007
or greater. In this subsample control group, 33 had had a previous diagnosis or drug treatment for anxiety or depression (Table 4). So the projected number of controls with a previous record of anxiety or depression would be 123 in the whole survey low-score group (n 1684). Our definition of population outcome is as follows: subjects in the case groups with a prior record of anxiety and depression represent those who had not fully recovered, and subjects in the low-score group with a prior record represent those who had recovered. The numbers in the study that were identified with such records were: 143 in the highscore group, 82 in the medium-score group, and an estimated 123 in the noncase group. This means that, among all survey responders with a diagnosis or drug treatment for anxiety or depression in the previous year (projected total n 143 82 123 348), 65% (143 82 out of 348) had not fully recovered, and an estimated 35% (123 out of 348) had recovered. 3.2.4. Responders compared with nonresponders There were 1362 (34%) people who did not respond to the questionnaire survey. A randomly selected sample (n 100) from this group was evaluated separately to investigate nonresponse bias. There was little gender difference between the nonresponder and responder groups, but the nonresponders were younger and more likely to be from deprived areas. The record review of the 12-month period
Table 2 Retrospective comparison of GP contact, diagnosis, and drug treatment by study groups Clinical record GP contact Diagnosis or treatment Diagnosis Treatment a
c
No Yes Noc Yes Noc Yes Noc Yes
Controls
Medium score
High score
ORa (95% CI)
ORb (95% CI)
96 (21.3) 354 (78.7) 417 (92.7) 33 (7.3) 429 (95.3) 21 (4.7) 428 (95.1) 22 (4.9)
74 (14.6) 432 (85.4) 424 (83.8) 82 (16.2) 455 (89.9) 51 (10.1) 445 (87.9) 61 (12.1)
39 (9.4) 377 (90.6) 273 (65.6) 143 (34.4) 297 (71.4) 119 (28.6) 305 (73.3) 111 (26.7)
1.0 1.5 (1.1–2.1) 1.0 2.3 (1.5–3.5) 1.0 2.2 (1.3–3.7) 1.0 2.4 (1.4–4.0)
1.0 2.5 (1.7–3.7) 1.0 6.4 (4.3–9.7) 1.0 8.0 (4.9–13.1) 1.0 6.9 (4.2–11.2)
Odds ratio for medium-score group versus control group adjusted for age, gender, and Townsend deprivation score. Odds ratio for high-score group versus control group adjusted for age, gender, and Townsend deprivation score. c Reference category. b
1116
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
Table 3 Time interval (days) from first record of anxiety or depression to survey, by study groups Clinical record
Group
Sample (n)
Median (days to survey)
Quartiles (lower, upper)
Range (min, max)
Diagnosis or treatment
Controls Medium-score High-score Controls Medium-score High-score Controls Medium-score High-score
33 82 143 21 51 119 22 61 111
288 283 279 288 252 266 283 265 265
196, 345 159, 335 133, 337 179, 344 117, 330 119, 326 204, 349 135, 334 109, 341
39 to 365 7 to 364 1 to 365 39 to 365 13 to 364 1 to 365 39 to 365 7 to 364 1 to 365
Diagnosis
Treatment
before the survey for the nonresponder sample showed that 70% had had a GP contact and 12% had a record of diagnosis or drug treatment for anxiety or depression. The only significant difference in the record review for the nonresponders group compared with the responder group was that they had had less GP contact, but in terms of recorded history of diagnoses and treatment for anxiety or depression, there were no differences (Table 5). Applying the results of the record review to all 1362 nonresponders, the projected number of nonresponders with a previously recorded diagnosis or drug treatment for anxiety or depression is estimated to be 163. If we repeat the calculations of population outcome for the whole surveyed population, including nonresponders, the figures would be as follows: 511 in total with a relevant record; 123 (24%) patients who had recovered; 225 (44%) who had not fully recovered, and 163 (32%) patients whose recovery was not determined. 4. Discussion 4.1. Study findings: population survey as an estimate of health care need The questionnaire survey confirmed the high prevalence of anxiety and depression symptoms in the adult UK population. Figures for the high score group were comparable to those of other studies [1,2], which have estimated the prevalence of neurotic disorders such as anxiety and depression at approxi-
mately 15% in the population. The associated demographic characteristics also confirm results from other studies [1,2]. 4.2. Study findings: population survey as an estimate of health care outcome Patients with a high HAD score were more likely to have contacted the GP for any reason, at least once, in the prior 12 months than those with milder or no symptoms. This group also had the highest proportion of patients with previous diagnosis or drug treatment for anxiety and depression. Previously diagnosed patients had been identified a median of 9 months before the survey, and so clinical outcome as measured by the questionnaire survey did not represent recent onset. One interpretation of this result is that, regardless of any interventions between diagnosis and the subsequent survey, this group represents poor outcome despite diagnosis. Because most of these diagnosed patients had received a drug treatment, this group also represents poor outcome despite treatment. While the score of the high-score group indicates poor outcome for previously diagnosed or treated patients, the noncase score for previously identified patients indicates “recovered” cases. Projections from our study suggest that, of all those in the practice population who had been diagnosed or treated for anxiety or depression in the course of 12 months, an estimated 35% had fully recovered, compared with the 65% who had not improved. This means that one-
Table 4 Estimated population outcome for surveyed patients with a prior record (diagnosis or treatment) of anxiety or depression Survey responders Controls n 1684 Medium-score n 506 High-score n 416 Total n 2606 a
Sampled records
Prior record of anxiety or depression in sampled records
Estimated number with record in all survey responders (%; 95% CI)
Relative outcomea
450
33
123b (4.7; 3.2 to 6.3)
123/348 35%
506
82
82 (3.2; 2.5 to 3.8)
82/348 24%
416
143
143 (5.5; 4.8 to 6.2)
143/348 41%
1372
258
348
100%
Estimated number of patients in the three separate outcome groups (as defined by their survey score) expressed as a percentage of all patients with a prior record of anxiety or depression (n 348). b Estimated total number of controls with a prior record of anxiety or depression, calculated by applying the results of the sampled records i.e. 33/450 (7.3%) to the total of 1684.
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
1117
Table 5 Comparison of sociodemographic and medical record characteristics between responders and nonresponders to the survey
Age
Gender Townsend score
GP contact Diagnosis or treatment Diagnosis Treatment
18 to 29 30 to 44 45 to 59 60 to 75 Female Male 1.00 0.99 to 0.01 0.00 to 0.99 1.00 No Yes No Yes No Yes No Yes
Respondersa % (95% CI)
Nonrespondersb % (95% CI)
16.9 (14.4–19.3) 27.7 (24.8–30.5) 28.4 (25.6–31.2) 27.1 (24.2–29.9) 54.1 (50.9–57.3) 45.9 (42.7–49.1) 14.6 (12.3–16.9) 30.9 (27.9–33.9) 42.7 (39.6–45.8) 11.0 (9.0–13.0) 18.1 (15.6–20.7) 81.9 (79.3–84.4) 86.6 (84.8–88.5) 13.4 (11.5–15.2) 90.5 (88.9–92.0) 9.5 (8.0–11.1) 90.2 (88.7–91.8) 9.8 (8.2–11.3)
35.0 (25.7–44.3) 34.0 (24.7–43.3) 19.0 (11.3–26.7) 12.0 (5.6–18.4) 48.0 (38.2–57.8) 52.0 (42.2–61.8) 6.2 (1.4–11.0) 27.8 (18.9–36.8) 55.7 (45.8–65.6) 10.3 (4.3–16.4) 30.0 (21.0–39.0) 70.0 (61.0–79.0) 88.0 (81.6–94.4) 12.0 (5.6–18.4) 92.0 (86.7–97.3) 8.0 (2.7–13.3) 92.0 (86.7–97.3) 8.0 (2.7–13.3)
a Percentage figures are based on extrapolation of results derived from the total record review sample of 1372 to the 2606 responders in the survey (using weighted linear function). b Percentage figures are based on the 100 nonresponders randomly sampled from the total of 1362 nonresponders.
third of anxious or depressed patients in the community had had their needs for health care identified in primary care but had also had those needs met effectively. The two-thirds with poor outcome included those who had been treated with drugs in the previous twelve months and yet were still cases in the survey. This may indicate a need for better use of drug treatment [19] or for more suitable and effective interventions. The relative proportions of those who had improved or not only relates to patients who were consulters and excludes patients who had not contacted their GP. The implications for health care of this group of nonconsulters who had had no record of practice contact might concern, for example, issues such as access to care, rather than outcomes of care. There is evidence from other sources to support the finding that, despite identification in general practice of patients with anxiety and depression; these patients may not improve in the long term. In a study comparing depressed patients whose status was disclosed to GPs with patients whose condition was not disclosed, disclosure led to worse clinical outcomes [22]. In an international study on outcome, over half of those detected to be depressed were still depressed 12 months later [3]. In using effective interventions, one may need to consider that patients with anxiety or depression may have a chronic or inevitably relapsing and recurrent condition or a more severe illness. The findings for patients in a medium HAD score group are more open to interpretation. The clear difference was that a higher proportion of this group had had a GP contact compared with the control group. However, the proportion with an anxiety or depression diagnosis or related drug treatment was lower than that in the high-score group. Thus, the medium-score score for those who were previously diagnosed
and treated patients could represent either a partially “poor” outcome (compared with controls) or a partially “good” outcome (compared with the high-score group). 4.3. Study findings: limitations in interpretation The categorization of the HAD score into high and low scores as a screening instrument allows the differentiation between those who were likely to be “true” cases as measured by high score and those who were “true” noncases as measured by a low score. In terms of the study, this categorization was used to estimate the outcomes for previously diagnosed patients, but does not necessarily imply that survey-defined “caseness” and clinically defined “caseness” are the same [12]. The possibility of censorship limits the interpretation of the estimated population outcomes based on the case and control groups. First, patients who were lost to follow-up as a result of anxiety or depression could potentially contribute to a censorship bias. One example of such a bias, suicidal death, was considered, and no such deaths were identified in the study period. Second, there was a 34% nonresponse rate to the survey, and the demographic comparisons showed that responders and nonresponders were different. However, based on our projections (see Results section), even if all 163 nonresponders with a previous diagnosis or treatment of anxiety or depression had recovered by the time of the survey, at least 44% of previously diagnosed subjects will still not have fully recovered. 4.4. Cross-sectional survey as an estimate of outcome: critique and potential There are three general cautions if our approach is to be used more widely. First, the issue of censorship bias, which
1118
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119
in the classic case-control study relates mainly to mortality. If the exposure under investigation causes death, the crosssectional study may underestimate any association between exposure and disease. For primary care epidemiology this may not often be an issue because many clinical outcomes relate to morbidity as opposed to mortality. However, nonresponders to the survey may also give rise to a similar bias. Second, cross-sectional surveys tend to pick up chronic cases, so that consultation and treatment for a 12-month period may reflect a longer period of ineffective or intermittent treatment, and so may be more about chronicity than about other acute or episodic problems. However, the control group provides a balancing estimate of beneficial clinical outcomes of both short- and longer-term problems. Third, the clinical outcome cannot be disentangled from potential unknown confounding factors. In our study, clinical outcome may have been related to unknown confounding factors other than treatment, although the results were adjusted for the important potential confounders of age, sex, and deprivation. This study was completed in a practice that had recorded all patient practice contacts and used standard diagnostic and drug treatment classification. Previous studies have validated the use of this type of computer data in general practice [23–25]. The clinical data were not standardized for the purposes of research but nevertheless captured the patient’s condition at time of contact as the general practitioner perceived it in actual practice. Therefore, all recorded clinical information has been accepted to possess “face validity.” The study highlights the advantages of well-organized, complete, and routinely collected general practice data. Cross-sectional population surveys are routinely used to estimate the need for health care, and yet have not been used for estimating outcome of health care by linking the survey to retrospective records. Our design paralleled the classic aetiological case-control study by being a cross-sectional study of current disease status, onto which we grafted an historical study of exposure. So the strengths of our method were its potential for practical application in local studies, and using a study population that was unselected and whose source of information was also unselected, because the general practitioner is the major gateway to all forms of health care in the UK. Our approach may provide a more realistic picture of the outcome of actual health care as it is provided and taken up within one general practice population, than one gleaned from the usual study of single interventions in selected populations. In conclusion, by combining the results of a cross-sectional survey with prior recorded clinical history, not only can health care needs be identified, but also the current effectiveness with which such needs may have been met can be estimated. There are caveats to the interpretation of results from this type of research method, and issues of selectivity of “chronic survivors” in the case group need to be addressed when considering generalizability. However, as a research method for primary care, the questionnaire survey
can thus not only be a static tool for estimating health care need but, by adding a time dimension from practice records, it can serve as a dynamic tool for providing a perspective on the outcome of such identified health needs. Acknowledgments North Staffordshire Health Authority supported UTK in a Public Health post during the project year and the Royal College of General Practitioners Scientific Foundation Board Grant funded part of the project. We are very grateful to all the patients and staff of the study practice. We would also like to thank Rob McCarney for his help in the survey data collection, Rhian Hughes for assistance in record data collection, and Paul Trinder for his comments on the earlier drafts of the paper. References [1] Robins LN, Regier DA. Psychiatric disorders in America. The Epidemiological Catchment Area Study. New York: Free Press, 1991. [2] Meltzer H, Gill B, Petticrew M, Hinks K. The prevalence of psychiatric morbidity among adults living in private households. OPCS Surveys of Psychiatric Morbidity in Great Britain: Report No.1. London: HMSO, 1995. [3] Goldberg D, Privett, Ustun B, Simon G, Linden M. The effects of detection and treatment on the outcome of major depression in primary care: a naturalistic study in 15 cities. Br J Gen Pract 1998;48:1840–4. [4] Melfi CA, Chawla AJ, Croghan TW, Hanna MP, Kennedy S, Sredl K. The effects of adherence to antidepressant treatment guidelines on relapse and recurrence of depression. Arch Gen Psychiatry 1998;55: 1128–32. [5] Clarke EA, Anderson TW. Does screening by “PAP” smears help prevent cervical cancer? A case-control study. Lancet 1979;2:1–4. [6] Jick H, Dinan N, Rothman KJ. Oral contraceptives and non-fatal myocardial infraction. JAMA 1978;239:1403–5. [7] Office of Population Census and Surveys (OPCS). Morbidity Statistics from General Practice. Fourth National Study 1991–1992. Series MB5 No. 3. London: HMSO. [8] VAMP Vision. Practice health systems limited. London; 1996. [9] Zigmond A, Snaith RP. The Hospital Anxiety and Depression Scale. Acta Psychiatr Scand 1983;67:361–70. [10] Wilkinson MJB, Barczak P. Psychiatric screening in general practice: comparisons of the General Health Questionnaire and the Hospital Anxiety and Depression Scale. J R Coll Gen Pract 1988;38:311–13. [11] Lewis G, Wessely S. Comparison of the General Health Questionnaire and the Hospital Anxiety and Depression Scale. Br J Psychiatry 1990;157:860–4. [12] Herrmann C. International experiences with the Hospital Anxiety and Depression scale—a review of validation data and clinical results. J Psychosom Res 1997;17–41. [13] Carroll BT, Kathol RG, Noyes R, Wald TG, Clamon GH. Screening for depression and anxiety in cancer patients, using the Hospital Anxiety and Depression Scale. Gen Hosp Psychiatry 1993;15:69–74. [14] Sartorius N, Ustun TB, Lecrubier Y, Wittchen H-U. Depression comorbid with anxiety: results from the WHO study on psychological disorders in primary health care. Br J Psych 1996;168(suppl 30):38–43. [15] Dean SG, Dean JA, Coulombier D, Brendel KA, Smith DC, Burton AH, Dicker RC, Sullivan K, Fagan RF, Arener TG. Epi-Info, Version 6: a word processing, database, and statistics program for epidemiology on microcomputers. Atlanta: Center for Disease Control and Prevention; 1994. [16] Townsend P, Phillimore P, Beattie A. Health and deprivation: Inequality and the North. London: Croom Helm, 1988.
U. T. Kadam et al. / Journal of Clinical Epidemiology 54 (2001) 1112–1119 [17] NHS Centre for Coding and Classification. Read Codes and the Terms Projects: a Brief Guide. London: DoH, 1993. [18] British Medical Association and Royal Pharmaceutical Society of Great Britain. British National Formulary. London: British Medical Association/The Pharmaceutical Press, 1996. [19] Donoghue JM, Tylee A. The treatment of depression: prescribing patterns of antidepressants in primary care in the UK. Br J Psychiatry 1996;168:164–8. [20] Ohayon MM, Caulet M, Priest RG, Guilleminault C. Psychotropic medication consumption patterns in the UK general population. J Clin Epidemiol 1998;51:273–83. [21] Statistical Package for the Social Sciences (SPSS) Version 9.0. Chertsey (UK): SPSS Inc., 1998.
1119
[22] Dowrick C, Buchan I. Twelve month outcome of depression in general practice: does detection or disclosure make a difference? BMJ 1995;311:1274–6. [23] Wilson AE, Pollock C, Weekes T, Dowell A. Can general practices provide useful information? Evaluation of primary health care information project in Northern England. J Epidemiol Community Health 1995;49:227–30. [24] Jick H, Jick SS, Derby LE. Validation of information recorded on general practitioner based computerized data resource in the United Kingdom. BMJ 1991;302:766–8. [25] Pringle M, Ward P, Chilvers C. Assessment of the completeness and accuracy of computer medical records in four practices committed to recording data on computer. Br J Gen Pract 1995;45:537–41.