Volume 174 Issue 3

Research Selective versus universal screening for gestational diabetes mellitus: an evaluation of predictive risk factors Richard X Davey and P Shane Hamblin MJA 2001; 174: 118-121 For editorial comment, see Wilson Abstract - Methods - Results - Discussion - Acknowledgements - References - Authors' details - - More articles on Obstetrics & gynaecology and women's health Abstract Objective: To assess whether selective screening for gestational diabetes mellitus (GDM) on the basis of risk-factor assessment is a practicable alternative to universal screening. Design: Case-control study. Setting: A 212-bed regional specialist hospital in Melbourne, providing services in obstetrics and gynaecology, paediatrics, geriatrics and rehabilitation. Subjects: 6032 women who gave birth at the hospital, May 1996 to August 1997 and November 1997 to August 1998; all were screened for GDM, and 313 were diagnosed with the condition. Main outcome measures: Odds ratios (ORs) for risk factors (age, obesity, family history of diabetes mellitus and high-risk racial heritage) in women with GDM compared to those without GDM; proportion of women with GDM whose diagnosis would have been missed by selective screening. Results: ORs were 1.9 for age ≥25 years (95% CI, 1.3-2.7), 2.3 for body mass index ≥27 kg/m2 (95% CI, 1.6-3.3), 2.5 for high-risk racial heritage (95% CI, 2.0-3.2), and 7.1 for family history of diabetes mellitus (95% CI, 5.6-8.9). Other proposed criteria (previous GDM and glycosuria) added no further diagnostic power. Selective screening using the above four criteria would have missed two of 313 cases (0.6%) and could have saved screening up to 1025 women without GDM (17% of all women). Conclusions: Selective screening for GDM based on prior risk assessment can reduce the need for testing, with negligible loss of diagnostic efficiency. Gestational diabetes mellitus (GDM) is officially described as carbohydrate intolerance with onset or first recognition during pregnancy.1 It is associated with increased incidence of maternal hypertension, pre-eclampsia and obstetric intervention; a third of women with GDM develop diabetes mellitus in later life. Babies of mothers with GDM may be either macrosomic or small-for-gestational-age, and may suffer birth trauma, hypoglycaemia and other metabolic disturbances. Potential effects later in the child's life are still debated. In Australia, at least 5% of pregnancies are affected by GDM. However, there is no sharply defined maternal blood glucose level beyond which morbidity invariably ensues in either mother or baby,2 and there is disagreement about how to diagnose GDM and how aggressively to treat it.3In 1998, both the American Diabetes Association (ADA) and the WHO Consultation on diabetes published recommendations on diagnosis and classification of diabetes mellitus that included comments on GDM.4,5 The Australasian Diabetes in Pregnancy Society (ADIPS) has also published GDM guidelines.6 Both the ADA and ADIPS recommendations acknowledge that there are variable levels of risk for GDM and, consequently, that selective rather than universal screening can be considered. Selective screening both reduces costs and, for women deemed not to need screening, eliminates the minor physical inconvenience of the procedure and any anxiety raised by the possibility of suffering diabetes. Both ADA and ADIPS list risk factors for GDM (Box 1). The WHO Consultation's delineation of risk factors for GDM was less clear.5 We tested the hypothesis that selective screening for GDM is a practicable alternative to universal screening. We also investigated the effect of using the different age criteria of ADIPS and ADA as a basis for selective screening. Methods We undertook a case-control study to compare the likelihood of particular risk factors (defined in Box 2) among women with and without GDM. We also determined the proportion of women with GDM whose diagnosis would have been missed by selective screening, based on different sets of risk factors. Study population Sunshine Hospital is a 212-bed regional specialist hospital in Melbourne, Victoria, which provides service in obstetrics and gynaecology, paediatrics, geriatrics and rehabilitation. The study population comprised all 6032 women who gave birth at the hospital over the 26 months May 1996 to August 1997 and November 1997 to August 1998. Women who gave birth in September and October 1997 were excluded, as their laboratory data were incomplete. All women were screened with a 50 g glucose challenge test, according to the ADIPS protocol.6 Those with an abnormal result (defined as plasma glucose level after one hour of ≥7.8 mmol/L) proceeded to a 2 h 75 g oral glucose tolerance test (OGTT); GDM was diagnosed if the fasting plasma glucose level was ≥5.5 mmol/L, or the 2 h level was ≥8.0 mmol/L. Nearly all patients diagnosed with GDM were managed by an endocrinologist (P S H) in conjunction with one of the clinic obstetricians and were offered review and ongoing care from a dietitian and a diabetes nurse educator. Data retrieval Case group: We identified all post-delivery patient separations coded for GDM by computer search of the hospital medical information system, with cross-referencing to laboratory, dietitian and diabetes nurse educator records. Women who gave birth twice in the study period were included only once, using details from their first GDM-affected pregnancy. There were 313 women diagnosed with GDM. Information on risk factors for these women was obtained from medical records containing details of pregnancy management and delivery, endocrinologist's and dietitian's notes and laboratory records (by R X D). If racial heritage was unclear, patients were telephoned at home to obtain more details. Body mass index (BMI) was available for only 290 of the 313 women (93%), but other data were available for over 99%. Control group: For the 5719 women without GDM, information on age was also obtained from the hospital medical records. However, it was impracticable to investigate racial heritage as closely for this group as for the case group. Therefore, if country of birth was recorded as being in Europe, Asia or Central and South America, it was used for risk categorisation (45.5% of women). All other women were allocated to risk groups in the same proportions as found in the case group. While this biases the outcome in favour of the null hypothesis, it is more accurate than making no such allocation at all. BMI and family history were not available for the 5719 women without GDM. Therefore, the BMI comparison used BMI data obtained from 303 consecutive non-diabetic women presenting for a glucose challenge test at about 28 weeks' gestation as part of a 1995 study at Sunshine Hospital.8 As the patient catchment area was unchanged between 1995 and 1998, this group should represent an unbiased sample of women who presented between 1995 and 1998. For the family history comparison, a recent estimate of prevalence of diabetes mellitus in Australia9 was used for the non-GDM patients. Background risk was corrected for the bias caused by the tendency of patients with diabetes to visit their doctors twice as often as non-diabetic patients.10 It was also doubled to give a worst-case estimate, as each parent might pass on heritable risk independently. Statistical analyses and ethics approval Data were analysed using Stata statistical software.11 Odds ratios were calculated from the comparative prevalence in affected and control populations by Cornfield's method. Ethics approval for this study was not required by the Victorian Health Services Act 1988 and was not sought. Data were permanently de-identified after analysis. Results Risk-factor comparison Prevalence of risk factors among women with and without GDM is shown in Box 3, along with odds ratios. Women with GDM were almost twice as likely to be aged 25 years or over compared with those without GDM, more than twice as likely to have a BMI ≥27 kg/m2 or to have a high-risk racial heritage, and more than seven times as likely to have a family history of diabetes mellitus. To determine the value of racial heritage as a predictor of GDM in isolation from family risk, we determined the OR for high-risk racial heritage among women with no family history of diabetes mellitus. This OR was not statistically different from the earlier OR for high-risk racial heritage that included women with a family history. Furthermore, birth in Australia, New Zealand or North America (of non-Indigenous background) does not necessarily equate with low heritable GDM risk. Of the 313 women with GDM, 94 were born in these countries, but 19 of these had high-risk racial heritage. Finally, we also assessed whether glycosuria in pregnancy or previous GDM had any extra value as predictors of GDM. All women with these risk factors qualified for screening on other grounds. Effect of selective screening The numbers of women with GDM who would be screened on the basis of risk factors, using different age thresholds, are shown in Box 4. Only the 290 women with complete data for all risk factors are included. However, all 23 women with incomplete risk-factor data would have undergone screening under these selective screening policies, as all had at least one risk factor (10 had two factors and five had three). Selective screening on the basis of at least one risk factor, using the ADIPS age criterion (≥30 years), would have missed 12 women with GDM (95% CI, 6-19; 4%). Using the ADA age criterion (≥25 years), selective screening would have missed only two women with GDM (95% CI, 0-5; 0.6%). Furthermore, χ2 tests showed that the proportions of women with GDM who had risk factors other than age did not differ significantly between age groups (≥30 years, ≥25 years and all ages); all P values exceeded 0.67. Among the women without GDM, 83% were aged 25 years or over and 48% were aged 30 years or over. A selective screening policy could therefore have saved testing up to 17%-52% of women without GDM, depending on the age threshold used and the presence of risk factors other than age. Discussion We found that selective screening for GDM using the four criteria common to the ADA and ADIPS lists of risk factors -- older age, obesity, family history of diabetes and high-risk racial heritage -- would have missed few women with GDM in our study population. It is clear that the age threshold for screening proposed by ADA (25 years) is diagnostically safer than the ADIPS threshold of 30 years, missing only 0.6% versus 4% of women with GDM. However, our data also show that it is important to examine risk factors other than age even when the lower age threshold is used, as the other factors underlying susceptibility to GDM operate irrespective of age. Family history and heredity are immutable, and obesity may also be partly under genetic control. These observations are also consonant with the theory that pregnancy unmasks diabetes mellitus prematurely.12 We also found that previous GDM and glycosuria in pregnancy added nothing to the above four criteria for screening. However, this does not mean that GDM in a previous pregnancy should be ignored. It is often regarded as a criterion for a full OGTT, without a prior glucose challenge test, earlier than 28 weeks' gestation; the wisdom of this practice is not doubted. Not only are the ADA criteria for screening diagnostically safer than the ADIPS criteria, they are also more precise in their definitions. Australian women would benefit if ADIPS recommendations were brought into line with ADA recommendations. Our conclusions differ from those of Moses and colleagues, who found no benefit from selective screening in their study in the Illawarra region of New South Wales.13,14 Indeed, Moses has championed universal screening.15 However, the Illawarra protocol for GDM screening varied from contemporary practice, as it did not measure fasting glucose level or stringently control the time between the glucose load and blood sampling, making comparison difficult. Its outcomes have been questioned.16,17 In North America, a recent, albeit small, retrospective study of GDM screening in Michigan specifically assessed the ADA selective screening recommendations and concluded that they can be used as they miss "few" (4%) women with GDM.18 A larger study was reported by the Toronto Trihospital Investigators.19 They proposed a scheme that used among its criteria those later published by ADA to differentiate risk levels for GDM, sparing 35% of pregnant women the need for a glucose challenge test. This is at least twice as efficient as using the ADA criteria in our population, which potentially spared up to 17% of women a glucose challenge test (depending on the presence of risk factors other than age). However, an accompanying editorial concluded that the Trihospital criteria were "so hard to discern" that universal screening would continue as the only practicable alternative.20 For busy clinicians, simple systems are essential. Our study differed from the Toronto study in that only women with positive results on a glucose challenge test proceeded to an OGTT, while, in Toronto, all women had a full OGTT. We will have missed the small number of women who would have had positive results on an OGTT despite their negative results on a challenge test -- perhaps 3%, based on the Toronto data. Short of performing a full OGTT on all pregnant women, which is impracticable, this group will always escape detection. The Toronto ORs for risk factors among those with GDM generally accord with ours (1.6 for age ≥35 years [95% CI, 1.1-2.5], 3.2 for BMI ≥25.1 kg/m2 [95% CI, 2.1-4.8], 4.8 for Asian race [95% CI, 3.0-7.6]), but our results differ in two ways. Toronto race groupings, apart from "Asian", are difficult to interpret and, by using "white" and "black", ignore the extreme variation among "white" Europeans. Secondly, family history of diabetes mellitus in Toronto did not correlate significantly with higher risk of GDM, whereas our findings strongly support the inclusion of family history among criteria for a glucose challenge test. Our study, along with the Michigan and Toronto studies, indicates that a selective approach to GDM screening in pregnancy is justifiable. Our simplified algorithm for selective screening is shown in Box 5. In practice, the proportion of women spared a glucose challenge test by selective screening will vary between populations. Consequently, whether selective screening is locally practicable will be a decision for individual groups of obstetricians, endocrinologists and pathologists with local knowledge. It is clear that consideration of a patient's age, rigorous questioning about racial and family history and accurate measurement of height and weight can reduce the need for screening among suitable populations. Selective screening can reduce costs and maternal anxiety, with negligible loss in diagnostic power. Nevertheless, GDM poses still further challenges. Australia needs a better-directed, more organised, totally inclusive approach to follow-up of women who have had GDM. The third who will go on to develop diabetes mellitus need to be tracked, monitored, and managed prospectively into a healthier future. Acknowledgements We thank dietitians Candy d'Menzie-Bunshaw and Ruth Cuttler, specialist diabetes nurse consultant Elizabeth Borg, and health information manager Sianne Banks and her staff at Sunshine Hospital for their invaluable assistance with the study, and Lucy Inocencio for her excellent technical assistance. References Metzger BE, editor. Summary and recommendations of the Third International Workshop-Conference on Gestational Diabetes Mellitus. Diabetes 1991; 40 Suppl 2: 197-201. Sacks DA, Greenspoon JS, Abu-Fadil S, et al. Towards universal criteria for gestational diabetes: The 75-gram glucose tolerance test in pregnancy. Am J Obstet Gynecol 1995; 172: 607-614. Jovanovic L. A tincture of time does not turn the tide [editorial]. Diabetes Care 2000; 23: 1219-1220. The Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Report of the expert committee on the diagnosis and classification of diabetes mellitus. Diabetes Care 1998; 21 (Suppl 1): S5-S19. Alberti KGMM, Zimmet PZ for the WHO Consultation. Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1. Diagnosis and classification of diabetes mellitus. Provisional report of a WHO consultation. Diabet Med 1998; 15: 539-553. Hoffman L, Nolan C, Wilson JD, et al, for the Australasian Diabetes in Pregnancy Society. Gestational diabetes mellitus -- management guidelines. Med J Aust 1998; 169: 93-97. Beischer NA, Oats JN, Henry OA, et al. Incidence and severity of gestational diabetes mellitus according to country of birth in women living in Australia. Diabetes 1991; 40: 35-38. Davey R. The glucose challenge test: different drink dilutions. Diabet Med 1996; 13: 917-918. Welborn TA, Reid CM, Marriott G. Australian Diabetes Screening Study: impaired glucose tolerance and non-insulin-dependent diabetes mellitus. Metabolism 1997; 46 (12 Suppl 1): 35-39. Australian Bureau of Statistics. National health survey: diabetes, Australia, 1995. Canberra: AGPS, 1997. (Catalogue No. 4371.0.) StataCorp. 1999. Stata Statistical Software: Release 6. College Station. TX: Stata Corporation. Yue DK, Molyneaux LM, Ross GP, et al. Why does ethnicity affect prevalence of gestational diabetes? The underwater volcano theory. Diabet Med 1996; 13: 748-752. Moses R, Griffiths R, Davis W. Gestational diabetes: do all women need to be tested? Aust N Z J Obstet Gynaecol 1995; 35: 387-389. Moses RG, Moses J, Davis WS. Gestational diabetes: do lean young Caucasian women need to be tested? Diabetes Care 1998; 21: 1803-1806. Moses RG. Diabetes in pregnancy [editorial]. Med J Aust 1998; 169: 68-69. Davey R. Of gestational diabetes, finesse, and an antipodean snark [letter]. Diabetes Care 1999; 22: 873-874. Moses RG, Moses J, Davis WS. Response to Davey [letter]. Diabetes Care 1999; 22: 874. Williams CB, Iqbal S, Zawacki CM, et al. Effect of selective screening for gestational diabetes. Diabetes Care 1999; 22: 418-421. Naylor CD, Sermer M, Chen E, Farine D. Selective screening for gestational diabetes mellitus. N Engl J Med 1997; 337: 1591-1596. Greene MF. Screening for gestational diabetes mellitus [editorial]. N Engl J Med 1997; 337: 1625-1626. (Received 15 May, accepted 21 Sep, 2000) Authors' details Western Hospital, Melbourne, VIC. Richard X Davey, FRCPA, FACB, Clinical Pathologist; P Shane Hamblin, FRACP, Senior Endocrinologist. Reprints will not be available from the authors. Correspondence: Dr R X Davey, Western Hospital, Gordon Street, Footscray, VIC 3011. richard.daveyATwh.org.au Make a comment 1: Risk factors for gestational diabetes mellitus, listed by different sources Risk factor Australasian Diabetes in Pregnancy Society6 American Diabetes Association4 Age Obesity Family history of diabetes mellitus Previous GDM High risk "ethnic" group Glycosuria Previous adverse pregnancy outcome Yes (>30 years) Yes (not defined) Yes Yes Yes (examples given)† Yes Yes Yes (>25 years) Yes (BMI >27kg/m2) Yes (first-degree relative) Not mentioned* Yes (examples given)‡ Not mentioned Not mentioned* GDM=Gestational diabetes mellitus. BMI=Body mass index. *While the ADA did not consider previous GDM or adverse pregnancy outcome as sufficiently significant for women to be included in the high-risk GDM group, it did report them as criteria for diabetes testing in asymptomatic, undiagnosed individuals,4 thereby acknowledging them as markers of early, silent diabetes mellitus. †Including Australian Indigenous, Polynesian, Asian and Middle Eastern women. ‡Including Hispanic-American, Native American, Asian-American, African-American and Pacific Islander women. Back to text 2: Definitions of risk factors used in this study Age: Age was not further defined by ADA or ADIPS; we used age at estimated time of conception — the most conservative calculation. Obesity: As defined by the ADA — body mass index ≥27kg/m2, determined from pre-pregnancy mass and height. Family history of diabetes mellitus: As defined by the ADA — diabetes mellitus affecting a first-degree relative. Racial susceptibility: Termed "ethnic" risk by ADA and ADIPS. We classified a woman as having high-risk racial heritage if she or her parents were born in one of the countries around the Mediterranean (including the Levant, but not the rest of Europe), the Indian subcontinent or Asia, or belonged to the Indigenous populations of Australia, the Pacific or the Americas.7 ADA = American Diabetes Association. ADIPS=Australasian Diabetes in Pregnancy Society. Back to text 3: Prevalence and odds ratios of risk factors for gestational diabetes mellitus (GDM) Prevalence Risk factor Women with GDM (n=313) Women without GDM (variable n*) Odds ratio (95% CI) Age (years) >25 90.1% 82.9% 1.9 (1.3-2.7) >30 58.5% 47.8% 1.5 (1.2-1.9) Body mass index >27 kg/m2 36.2%† 19.8% 2.3 (1.6-3.3) Family history of diabetes mellitus 39.9% 8.6% 7.1 (5.6-8.9) High-risk racial heritage 68.7%‡ 46.4% 2.5 (2.0-3.2) Among women with no family history 71.7%§ 42.8%¶ 2.9 (2.1-4.0) *Sample size varied between risk factors: 5719 (age), 303 (body mass index), 50371 (family history) and 5719 (racial heritage). †Data were available for 290 of the 313 women. ‡High-risk racial heritage: peri-Mediterranean (56 women), Indian subcontinent (20), Asia (124), South America (12), Indigenous populations (3); low risk racial heritage: United Kingdom (61), other European countries (35) and other (2). §n=187. ¶n=5324. Back to text 4: Effect if selective screening were used among 290* women with gestational diabetes mellitus Number of women who would be screened (% of women with GDM) Risk factors† ADIPS age threshold (≥30 years) ADA age threshold (≥25 years) Age ≥ threshold Only risk factor Plus any one other factor Plus any two other factors Plus any three other factors Total 13 (4%) 78 (27%) 54 (19%) 20 (7%) 165 (57%) 23 (8%) 120 (41%) 90 (31%) 27 (9%) 260 (90%) Age ≤ threshold One risk factor Two risk factors Three risk factors Total 60 (21%) 42 (14%) 11 (4%) 113 (39%) 18 (6%) 6 (2%) 4 (1%) 28 (10%) Any risk factor 278 (96%) 288 (99%) *23 women with gestational diabetes mellitus but incomplete data on risk factors are not included. †Risk factors other than age were body mass index ≥27kg/m2, family history of diabetes mellitus and high-risk racial heritage. Back to text Back to text

Richard X Davey

Abstract - Methods - Results - Discussion - Acknowledgement - References - Authors' details - - More articles on Administration and health services Abstract Objective: To quantify the morbidity and mortality associated with acute interhospital transfer of critically ill patients requiring intensive care (ICU) services. Design: Three-year (1 July 1996 - 30 June 1999) retrospective case-control study based on review of patients' medical records. Setting: Metropolitan hospitals in Melbourne, Victoria. Participants: 73 (of 75) consecutive, critically ill patients from one metropolitan teaching hospital who were transferred to other hospitals because ICU services were not available. Outcome measures: Primary endpoints included inhospital mortality and length of stay in ICU and hospital. Secondary endpoints included time from study entry to ICU admission and the change in predicted mortality risk after resuscitation and transfer to ICU (inter- or intrahospital transfer). Results: The Transfer Group experienced a significant delay in admission to ICU (5.0 [4.0-6.0] v 3.0 [2.0-5.5] hours; P = 0.001), and a longer stay in ICU (48 [33-111] v 44 [25-78] hours; P = 0.04), and hospital (10 [3-14] v 6 [3-13] days; P = 0.02). Hospital mortality in the Transfer Group (24.7%) was not statistically different from that in the Control Group (17.8%; P = 0.41; OR, 1.5; 95% CI, 0.68-3.4). Conclusion: Acute interhospital transfer is associated with a delay in ICU admission and a longer stay in ICU and hospital, but no statistically significant difference in mortality. A study of over 300 patient transfers would be required to clarify the morbidity and mortality risk of acute interhospital transfer. Acute interhospital transfer of critically ill patients carries potential risks, including complications during transfer and delay in providing definitive care. There are two categories of acute interhospital transfer. Category A: The primary (sending) hospital is unable to provide the expertise, diagnostic services or therapeutic procedures required by a patient (eg, transfer of a patient with extensive burns to a hospital with specialised treatment facilities for burns); and Category B: The primary hospital is temporarily unable to provide intensive care unit (ICU) services for a patient because of resource limitations (eg, lack of ICU beds). Measuring the impact of transfer risk on patient outcomes is complex. Reports without control data suggest that acute interhospital transfer increases both morbidity1,2 and mortality,3 but there are many confounding variables that influence outcome (eg, severity of illness, extent of resuscitation, and the expertise available before and during transfer). To our knowledge, no comparative outcome study of critically ill patients undergoing acute interhospital transfer has been published. A randomised trial of Category A transfer would be complicated by the difficulty of finding a clinically and ethically appropriate control group. An alternative is to compare outcomes of patients undergoing Category B transfer with outcomes of matched patients not transferred. Both these groups of patients are resuscitated and managed with similar expertise and support. It is therefore possible to reduce bias from some of the confounding variables, and to identify a suitable control group. We performed a retrospective case-control comparison of outcomes in critically ill adult patients undergoing Category B transfer. We hypothesised that acute interhospital transfer increases morbidity and mortality and sought to answer four questions: Does acute interhospital transfer of critically ill patients delay admission to an ICU; increase severity of illness before admission; increase ICU and hospital length of stay; and increase mortality? Methods The primary hospital was the Northern Hospital, a Melbourne metropolitan teaching hospital providing all acute-care health services (except cardiac surgery and organ transplantation).* Transfer Group All adult patients requiring intensive care services between 1 July 1996 and 30 June 1999 were entered in the study if they were deemed by the intensivist on-duty to require intensive care services; were transferred to a (public or private) metropolitan hospital for those services; and received diagnostic and therapeutic interventions that could otherwise have been provided at the primary hospital. All patients were transferred by road ambulance with an experienced medical escort from the primary hospital. Patients were excluded if they were transferred with the intention of receiving services not available at the primary hospital (Category A transfer), or if insufficient data were available. Control Group Control patients were selected from patients admitted to the ICU of the primary hospital, who did not undergo acute interhospital transfer at any time during the study period. Matching of Control Group patients with patients in the Transfer Group was undertaken according to a hierarchy of criteria deemed most likely to influence patient outcome (Box 1). Due to difficulties matching for all criteria, priority was placed on the first four. Matching was undertaken by one author (J V G), who was blinded to the identity and outcome of the patients. Endpoints These included inhospital mortality, time from study entry to ICU admission, length of stay in ICU and hospital, and the change in predicted mortality risk after resuscitation and transfer to ICU (Box 2). Patient data (Box 2) Patients' data recorded prospectively in the medical records at all the hospitals involved were reviewed retrospecively by one investigator (G J D). Physiological and laboratory data were used to calculate predicted mortality risk (pm) at three time points (t1, t2 and t3; Box 2), as an index of illness severity, using the Acute Physiology and Chronic Health Evaluation (APACHE) II method.4 Length of stay and pm were chosen as surrogate markers of patient morbidity. Because of the broad range of pm within the Transfer Group (0.01-0.86), we also undertook a post-hoc analysis of the change in pm (Δpm) during each interval as a measure of the change in physiological status. Since pm is an indicator of illness severity, and since calculations were performed before and after resuscitation and transfer, the physiological impact of resuscitation ([pm at t2] - [pm at t1]) and of transfer ([pm at t3] - [pm at t2]) was quantified. Statistical analysis Graph-Pad PRISM statistical package was used for data analysis.5 We used Fisher's exact test to compare group mortality. Non-parametric tests were used to compare length of stay in ICU and in hospital, and for intergroup comparison of pm and Δm (Mann-Whitney; P < 0.05). Wilcoxon signed rank test was used for post-hoc intragroup comparisons of Δm (P < 0.01). Data are presented as median (interquartile range) unless otherwise indicated. Based on studies without control data,3,6 which showed a doubling of mortality after acute interhospital transfer and a Control Group mortality of 18%, we calculated that a sample size of at least 50 patients would be required (α = 0.05, β = 0.80.) Ethical approval Ethics committee approval was obtained from each of the 14 hospitals involved. Results During the 36 months, 1470 patients required intensive care services at the primary hospital. Of these, 1338 (91%) were admitted and were the source of the Control Group patients. Of the 132 (9%) patients not admitted to the primary hospital ICU, 75 consecutive patients (5.1%) underwent a Category B transfer (Transfer Group), 35 (2.4%) were managed elsewhere within the same hospital (eg, general ward) and 22 (1.5%) were transferred for services not available at the primary hospital (Category A transfers). The last two groups were excluded from the study. Two eligible patients in the Transfer Group were excluded because insufficient data were available from the receiving hospitals, leaving 73 patients. The destinations of the transferred patients were determined by the proximity of the other hospitals and the bed availability. Sixty-four patients were transferred to nine public hospitals, and 11 patients were transferred to five private hospitals (eight of these patients had no private health insurance, but no public hospital ICU bed was available within the metropolitan region at that time). The reasons for transfer were closure of beds in 61 patients (84%), and equipment problems, all beds occupied and patient request in six, five and one patient, respectively. Demographic and other data for the Transfer and Control group patients, as well as the accuracy of case-control matching, are summarised in Box 3, and the diagnostic categories of the patients are shown in Box 4. Both groups had a high mortality risk at the time of study entry (commencement of resuscitation -- pm at t1) and immediately before transfer (pm at t2) (Box 5). Post-hoc analysis of Δpm revealed a significantly greater fall in pm during resuscitation and during transfer to ICU in the Control Group patients (P < 0.01). Thirty-seven patients (51%) undergoing acute interhospital transfer experienced a rise in pm (t3), compared with only 20 (27%) of the control group (P = 0.006). No deaths occurred during transfer. The higher observed mortality in the Transfer Group (Box 4) was not statistically different from that in the Control Group (odds ratio [OR], 1.5; 95% CI, 0.68-3.4). The diagnostic group (Box 4) and severity of illness (pm) were the most important univariate factors associated with outcome. Acute interhospital transfer was associated with a significant delay in ICU admission (Box 5), although some of the control patients also experienced admission delays (range, 0.5-9.5 hours). The Transfer Group was also found to have a significantly prolonged length of stay in ICU and hospital when compared with the Control Group. These length-of-stay increases were independent of outcome, diagnosis, age and hospital destination. Discussion We found that critically ill patients undergoing acute interhospital transfer experience a delay in admission to ICU, and a longer length of stay in ICU and hospital. However, there was no significant difference in hospital mortality between the two groups, and there were no deaths during transfer. As indicated by their primary diagnoses, need for life-support and high mortality risk, all patients in our study were critically ill at the time of study entry. The apparent safety of acute interhospital transfer is likely to be the combined result of factors such as resuscitation and stabilisation before transfer; the use of staffed and equipped ambulance vehicles; the provision of intensive care medical expertise and monitoring before, during and after transfer; and triage to appropriate hospitals.7 Why did the Transfer Group have an increased length of stay? The rise in pm after acute interhospital transfer in 37 patients (51%) suggests that it may increase morbidity in some patients. Other researchers have also reported adverse physiological effects during transfer of critically ill patients,1,2 and a higher mortality in patients admitted after interhospital transfer.3,6 Delay in ICU admission inevitably delays diagnostic and therapeutic procedures. The increased sedation and analgesia to ensure patient safety and comfort during interhospital transfer may prolong recovery time. The slower rate of fall of pm in the Transfer Group is consistent with this premise. Patient management and discharge practices may vary between institutions and thus increase length of stay independent of diagnosis and patient origin. Our study has several important limitations. Retrospective chart analysis carries potential for observer bias and systematic error. We attempted to minimise this by sampling data at predetermined fixed time points and using the same data collector. Patient selection was unavoidably biased because the Transfer Group constituted a heterogeneous and non-randomised group of critically ill patients. Because of the small number of subjects in some diagnostic categories, the matching of Control Group patients with patients in the Transfer Group was not perfect, but we attempted to optimise matching by using a criteria hierarchy. The Control Group patients had a greater median pm at study entry, and some experienced a clinically significant delay in ICU admission, both factors which may have reduced the outcome difference between the groups. Although the APACHE-II scoring system4 has been used in the prehospital setting,3,8 it assumes patients are in an intensive care environment receiving optimal therapy, and therefore it may not be a valid tool for use outside an ICU. However, this potential systematic error applied equally to both groups. The study size had insufficient power to establish a difference in mortality. If the observed difference in outcome is clinically significant it would require a sample size of over 300 transfers to exclude a type II statistical error. A trial of sufficient power could be achieved with a 12-month multicentre study of all Category B transfers within metropolitan Melbourne. During 1998-1999, 369 critically ill adults (3.5% of metropolitan adult intensive care admissions) underwent a Category B transfer (Department of Human Services, Critical Care Inter-Hospital Transfer Monitoring and Advisory Group, personal communication). At best, our results indicate that acute interhospital transfer does not affect hospital outcome; at worst, they suggest that it may adversely affect the outcome of one in every 25 critically ill patients transferred. Extrapolating our results to the metropolitan region, acute interhospital transfer may adversely affect the outcome of 15 patients (95% CI, 0-48) per annum, and require an additional 1100 hospital bed-days (95% CI, 960-1266 days) per annum -- half in ICU, where the primary resource limitation exists.9,10 Acknowledgement We would like to thank Professor B Jackson and Dr P Cranswick for their constructive criticism of the manuscript. References Waddell G, Scott PDR, Lees NW, Ledingham IM. Effects of ambulance transport in critically ill patients. BMJ 1975; 1: 386-389. Karipis H, Scheinkestel CD, Tuxen DV, et al. Safety of transportation of critically ill patients. Anaesth Intensive Care 1993; 21: A7111. Bristow P, Brown D, Lee A, Buist M. Transfer of severely ill patients. Anaesth Intensive Care 1995; 23: A399. Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med 1985; 13: 818-829. GraphPad PRISM, version 1. San Diego: GraphPad Software Inc, 1998. Metcalf A, McPherson K. Study of provision of intensive care in England, 1993. London: School of Hygiene and Tropical Medicine, 1995. Faculty of Intensive Care, Australian and New Zealand College of Anaesthetists and Australasian College of Emergency Medicine. Minimum standards for transport of the critically ill (IC-10). Melbourne: Australian and New Zealand College of Anaesthetists and Australasian College of Emergency Medicine, 1996. Bion JF, Edlin SA, Ramsay G, et al. Validation of a prognostic score in critically ill patients undergoing transport. BMJ 1985; 291: 432-434. Acute Health Services Branch, Department of Health and Community Services Review of emergency and critical care services in Victoria. Melbourne: Department of Health and Community Services, 1994. Acute Health Division, Department of Human Services. Review of intensive care in Victoria [Phase 1 report]. Melbourne: Department of Human Services, 1997. (Received 3 Apr, accepted 15 Sep, 2000) Authors' details Intensive Care Department, The Northern Hospital, Melbourne, VIC. Graeme J Duke, MB BS, FFICANZCA, Director; John V Green, MB BS, FFICANZCA, Staff Specialist. Reprints will not be available from the authors. Correspondence: Dr G J Duke, Intensive Care Department, The Northern Hospital, 185 Cooper Street, Epping, VIC 3076. graeme.dukeATnh.org.au 1: Criteria for matching Control Group patients with Transfer Group patients Discharge diagnosis (APACHE-III diagnostic code) Need for mechanical ventilation on admission to Intensive Care Unit Glasgow coma score (GCS) at t1 - within 2 points Predicted mortality (pm; APACHE-II methodology4) at t1 - within 10% Age - within 10 years Sex Source of initial referral (emergency ward, inpatient ward, operating theatre) Date of admission - within 12 months Time (t1): day (8:00 to 18:00) or night APACHE=Acute Physiology and Chronic Health Evaluation. Time, t1=study entry at commencement of resuscitation. Back to text 2: Patient dataset Demographic information, including age, sex, and postcode of residence Relevant medical data, including past history and final diagnosis Dates and times of primary admission, initial referral, interhospital transfer, ICU discharge and hospital discharge Referral source Treatment and personnel required during transfer Interventions required (at both hospitals) Data for APACHE II predicted mortality (pm) score4* Physiological data: blood pressure, heart rate, respiratory rate, Glasgow coma score, urine output Pathological data: haematocrit and total white cell count; serum levels of sodium, potassium, creatinine, urea, albumin, and glucose; and arterial pH and blood gas analysis Clinical data: age, diagnosis, use of mechanical ventilation, presence of acute renal failure, chronic health status Time of APACHE II predicted mortality (pm) calculations (see time line) t1= study entry at commencement of resuscitation. t2= before transfer to ICU, after initial resuscitation. t3= on arrival in ICU after transfer. APACHE=Acute Physiology and Chronic Health Evaluation. *Formula for pm score: logn(pm/12pm)=-3.517+0.146 (k1+k2+k3)+k4+k5, where k1=a variable score based on physiological and pathological data; k2=a variable weighting for age; k3=a variable weighting for chronic health status; k4=a constant weighting for emergency surgical patients; and k5=a variable weighting for principal diagnostic category. Back to text 3: Comparison of patient data (Transfer Group v Control Group - data are median and interquartile range unless indicated otherwise) and percentage matching between the two groups Criterion Transfer Group Control Group Percentage matching* Diagnostic group (see Box 4) (see Box 4) 100% Need for mechanical ventilation (no [%] of patients) 51 (73%) 51 (73%) 100% Predicted mortality at start of resuscitation (pm at t1) 0.30 (0.09-0.62) 0.37 (0.09-0.60) 69% GCS: patients with neurological problems (n=39) 7 (6-9) 7 (6-8) 100% GCS: all patients (n=73) 9 (6-14) 8 (6-12) 96% Age (years) 54.8 (36.4-67.2) 57.4 (38.0-70.4) 67% Source of referral (no [%] of patients) 61 (83%) from ED 59 (81%) from ED 79% Sex ratio (no. of men:women) 39:34 46:27 78% Date of admission (baseline) 3 (1-15) months 75% Time (t1) (8:00-18:00) (no. [%] of patients) 28 (38%) 36 (49%) 60% *Percentage of Control Group patients matched according to criteria given in Box 1. GCS=Glasgow Coma Score. ED=emergency department. Back to text 4: Diagnostic categories and inhospital mortality Deaths Discharge diagnosis No. (%) patients Transfer Group Control Group Trauma Drug overdose Cardiac arrest Cardiogenic shock Exacerbation of COPD Status asthmaticus Pneumonia Cerebrovascular coma Metabolic coma Neurological conditions Gastrointestinal conditions Septicaemia Malignancy Aortic aneurysm Total 12 (16%) 11 (15%) 8 (11%) 7 (10%) 5 (7%) 3 (4%) 4 (5%) 4 (5%) 4 (5%) 4 (5%) 4 (5%) 4 (5%) 2 (3%) 1 (1%) 73 (100%) 0 0 6 0 0 0 2 2 1 1 2 3 1 0 18 (24.7%) 0 0 5 1 0 0 1 1 2 1 1 1 0 0 13 (17.8%) 95% CI 11-25 7-20 COPD=Chronic obstructive pulmonary disease. Back to text 5: Acute Physiology and Chronic Health Evaluation (APACHE) II predicted risk of death (pm; median and interquartile range) and outcomes Transfer Group Control Group P* APACHE II risk of death pm at t1 (at study entry) 0.30 (0.09-0.62) 0.37 (0.09-0.60) 0.87 pm at t2 (before transfer) 0.24 (0.06-0.40) 0.25 (0.06-0.47) 0.69 pm at t3 (at ICU entry) 0.21 (0.06-0.46) 0.16 (0.04-0.46) 0.63 Outcomes Admission delay (t3 - t1; hours) 5.0 (4.0-6.0) 3.0 (2.0-5.5) 0.001 Length of stay ICU (hours) 48 (33-111) 44 (25-78) 0.04 Hospital (days) 10 (3-14) 6 (3-13) 0.02 Mortality (no. of patients) 18 (95% CI, 11-25) 13 (95% CI, 7-20) 0.41 ICU=Intensive Care Unit. * Mann-Whitney test. Back to text

Graeme J Duke · John V Green

A randomised controlled trial was planned to compare two different treatment strategies — structured problem solving and selective serotonin reuptake inhibitor (SSRI) medication — for patients with mild to moderate major depression. The trial was to be conducted in the primary care setting with all treatment given by general practitioners. When no patients had been recruited into the study after six months, we performed an audit of all patients with depressive symptoms attending the doctors' practices over three weeks. Exclusion criteria were changed to ease entry into the trial, but still no patients were recruited over the following six months. What went wrong? MJA 2001; 174: 144-146 Why did the trial fail? - Acknowledgements - References - Authors' details - - More articles on General practice and primary care The recent National Survey of Mental Health and Wellbeing found that depression was associated with significant disability and that 6.3% of the Australian population was estimated to have suffered a major depressive disorder in the previous 12 months.1 The survey also showed that general practices were the main points of contact for patients with a mental disorder, consistent with previous reports that only 5% of such patients are referred to psychiatrists.2 As most depressed patients will be treated in primary care, the evaluation of treatment in this setting is important. Structured problem solving is emerging as an effective treatment for clinical depression.3-6 This treatment aims to teach patients to use their own resources to deal with their problems and includes skills such as identifying and simplifying problems, "brainstorming" potential solutions, and implementing these solutions. The treatment is brief, has clearly identified stages, and is very suitable for delivery by primary healthcare professionals. However, little research in primary care settings has used general practitioners as the main treatment providers, and thus the degree to which the evidence for the effectiveness of structured problem solving can be generalised to general practice is questionable. For this reason we designed a randomised trial with all treatment conducted by GPs (Box 1). The study was, in part, a replication of an earlier United Kingdom project that showed that problem solving was better than placebo, but equivalent to amitriptyline.4 Six months after the trial commenced no patients had been recruited. Why did the trial fail? In practical terms, the trial should not have failed, as mild to moderate depression is a common presentation in primary care and the research protocol addressed many of the problems identified in previous primary care research failures.9,10 For example: We used GPs who had participated in a Masters program designed to improve the recognition and management of mental disorders in general practice; We involved the participating GPs in the development of the protocols; We made every effort to minimise tasks involved in the study for the GPs; and The chief investigator was responsible for the teaching on the Masters program, and had developed a close working relationship with the GPs. While our original intention was not to address the complexity of conducting randomised trials in clinical practice, the apparent unease of many of the GPs with randomisation raises some potential reasons for the failure (Box 2). Apparent ambivalence towards randomisation in medicine, despite strong support in the scientific literature, has been reported previously.11 Silverman argues that as medicine shifts from the traditional authoritarian stance of "doctor knows best" towards the use of clinical guidelines and standardised treatment protocols, there is an increased discomfort in any approach that might be seen to admit a lack of crucial knowledge on the part of the doctor.11 In other words, asking patients to consent to randomisation between two conditions admits an uncertainty that compromises the traditional doctor-patient relationship. However, it has not been difficult to engage doctors in trials involving randomisation of their patients, and there is evidence that many are prepared to follow simple randomisation protocols.12 For example, one of the first randomised controlled trials of giving aspirin to patients very early after myocardial infarction (to stop or reverse the thrombotic process) enrolled 2500 GPs. These GPs, who were blind to the treatment condition they were offering, agreed to give the allocated capsules (aspirin or placebo) to any patient who presented with chest pain or other symptoms likely to be caused by a myocardial infarct. Two thousand patients were recruited into the trial, suggesting that, in this example, randomisation was not a substantial concern. If it is not randomisation per se that causes reluctance to recruit, then other factors need consideration. In our study, the audit results indicated that one in six patients had been excluded because the GPs lacked confidence that structured problem solving would be of benefit. Perhaps the GPs lacked confidence in their ability to deliver the psychological treatment effectively. If this is the case, the failure of this trial may have been influenced by the inclusion of a psychological treatment. We have argued elsewhere that doctors remain cautious about using non-drug treatments, partly because of the lack of organised promotion of non-proprietary treatments, and partly because of the difficulty in ensuring quality control.13 Yet, these GPs had agreed problem solving was a useful treatment, had demonstrated competence as part of their training, and two GPs in the group had conducted their own research projects where they taught structured problem solving to other doctors. It is also possible that, given the obvious differences between the psychological and pharmacological approaches, the GPs had formed an opinion early in the recruitment process that one or other approach would better suit a particular patient. In this case, exclusion from the trial was based on the presumption that the doctor already knew what treatment was best.11,14 In a disturbing example of "doctor knows best", two-thirds of suitable patients were not enrolled in a randomised trial of antiarrhythmia drugs because their doctors were so convinced of the benefits of the drugs that they did not want their patients allocated to the placebo group.15 The study eventually showed that these drugs were capable of causing fatal arrhythmias, so those doctors were essentially withholding from their patients the 50% chance of being allocated to the safer placebo alternative. Other factors may cause doctors to hesitate when faced with recruiting patients, including the complexities of obtaining informed consent, the need to complete clinical ratings or ask patients to complete self-report questionnaires for measurement of outcome, or the need to work within a specific treatment protocol. Although tasks for the GPs were minimal, it is likely that the necessary role change from practitioner to scientist-practitioner was too great to facilitate a shift in treating behaviour. If this is the case, the inherent contradiction between research and delivery of care in the minds of many clinicians is a basic problem for clinical research.16 Nevertheless, treatments that are efficacious in research settings need to be evaluated under the conditions of routine care, and it may not be sufficient to rely on the use of qualitative methods to evaluate outcome in primary care.17 Perhaps an explicit use of the "uncertainty principle" in randomised controlled trials within routine clinical care will enhance recruitment rates.18 That is, if there is apparent certainty about what treatment is best, it is ethically untenable that patients should have their treatment chosen at random, so only patients for whom there is uncertainty about which treatment would be best should be recruited into a randomised trial. In this way the ethical dilemma for clinicians is solved, and the heterogeneity of the patient sample maintained, as clinicians will differ significantly in the types of patients they will be uncertain about. Financial incentives might increase the involvement of clinicians in research, but have caused public outcry in the United States;19,20 the resulting assertive recruiting can also erode informed consent.21,22 We argue instead that a more fundamental shift in ethos and knowledge of principles that underlie research in clinical practice is required. This argument has parallels in the recent Royal Australian College of General Practitioners report on an implementation strategy for evidence-based clinical practice guidelines.23 The report points to a lack of understanding of the principles underpinning the use of clinical practice guidelines in general practice, and the need to develop a culture in which such guidelines are used and valued. In regard to outcome research, until doctors accept a scientist-practitioner model of practice it is unlikely that they will feel comfortable using conventional research protocols. Clinical research conducted in routine care will help clinicians make informed decisions about what may be the best treatment for their patients based on scientifically derived knowledge. We hope that the questions raised in this article will stimulate debate and research that will directly address this important issue. Acknowledgements This research was supported by a grant from the School of Psychiatry, University of New South Wales. References Andrews G, Henderson S, Hall W. Prevalence, comorbidity, disability and service utilisation; and overview of the Australian national mental health survey. Br J Psychiatry 2001. In press. Gath D, Catalan J. The treatment of emotional disorders in general practice: psychological methods versus medication. J Psychosom Res 1986; 30: 381-386. Mynors-Wallis L, Davies I, Gray A, et al. A randomised controlled trial and cost analysis of problem-solving treatment for emotional disorders given by community nurses in primary care. Br J Psychiatry 1997; 170: 113-119. Mynors-Wallis LM, Gath DH, Lloyd-Thomas AR, Tomlinson D. Randomised controlled trial comparing problem-solving treatment with amitriptyline and placebo for major depression in primary care. BMJ 1995; 310: 441-445. Schulberg HC, Block MR, Madonia MJ, et al. Treating major depression in primary care practice. Eight-month clinical outcomes. Arch Gen Psychiatry 1996; 53: 913-919. Catalan J, Gath DH, Anastasiades P, et al. Evaluation of a brief psychological treatment for emotional disorders in primary care. Psychol Med 1991; 21: 1013-1018. Depression Guideline Panel. Depression in primary care. Vol. 2. Treatment of major depression. Clinical Practice Guideline No. 5. Rockville, MD: Department of Health and Human Services, Public Health Service, Agency for Health Care Policy and Research, April 1993. (AHCPR Publication No. 93-0551.) ICD-10 classification of mental and behavioural disorders. Geneva: World Health Organization, 1992. Foy R, Parry J, McAvoy B. Clinical trials in primary care. BMJ 1998; 317: 1168-1169. Peto V, Coulter A, Bond A. Factors affecting general practitioners' recruitment of patients into a prospective study. Family Practice 1993; 10: 207-211. Silverman W. Equitable distribution of the risks and benefits associated with medical innovations. In: Maynard A, Chalmers I, editors. Non-random reflections on health services research: on the 25th anniversary of Archie Cochrane's effectiveness and efficiency. London: BMJ Publishing Group, 1997: 184-193. Elwood P. Cochrane and the benefits of aspirin. In: Maynard A, Chalmers I, editors. Non-random reflections on health services research: on the 25th anniversary of Archie Cochrane's effectiveness and efficiency. London: BMJ Publishing Group, 1997: 107-121. Andrews G. On the promotion of non-drug treatments. BMJ 1984; 289: 994-995. Segelov E, Tattersall MHN, Coates AS. Redressing the balance -- the ethics of not entering an eligible patient on a randomised trial. Ann Oncol 1992; 3: 103-105. Moore TJ. Deadly medicine. New York: Simon and Schuster, 1995. Tognoni G, Alli F, Avanzini F, et al. Randomised clinical trials in general practice: lessons from a failure. BMJ 1991; 303: 969-971. Miller ML, Crabtree BF. Qualitative analysis: how to begin making sense. Family Practice Res J 1994; 14: 289-297. Collins R, Peto R, Gray R, Parish S. Large-scale randomised evidence: trials and overviews. In: Maynard A, Chalmers I, editors. Non-random reflections on health services research: on the 25th anniversary of Archie Cochrane's effectiveness and efficiency. London: BMJ Publishing Group, 1997: 197-230. Eichenwald K, Kolata G. Drug trials hide conflicts for doctors. New York Times, May 16, 1999; 1,28-29. Eichenwald K, Kolata G. A doctor's drug trials turn into fraud. New York Times, May 17, 1999. Ferguson C. Payment of financial incentives to GPs may invalidate informed consent process. BMJ 1998; 316: 75-76. Shalala D. Protecting research subjects -- what must be done. N Engl J Med 2000; 343: 808-810. Report on consultancy to develop an implementation strategy for CPGs. Sydney: Royal Australian College of General Practitioners, August 2, 2000. Authors' details School of Psychiatry, University of New South Wales, Sydney, NSW. Caroline J Hunt, MPsych, PhD, Lecturer (currently, Senior Lecturer, Department of Psychology, University of Sydney). Gavin Andrews, MD, Professor. Clinical Research Unit for Anxiety Disorders, St Vincent's Hospital, Sydney, NSW. Louise M Shepherd, BA(Hons), MPsych, Clinical Psychologist. Reprints will not be available from the authors. Correspondence: Dr C J Hunt, Department of Psychology (F12), University of Sydney, NSW, 2006. carolineATpsych.usyd.edu.au Make a comment 1: The planned trial Objective: To compare structured problem solving with SSRI medication and non-specific counselling for mild to moderate major depression. Trial development: The trial was initially designed with three treatment arms: structured problem solving, medication and placebo. In 1997, general practitioners in their second year of a two-year part-time Master of Psychological Medicine course were approached for feedback. The GPs indicated that, while they believed the study to be of value, they were uncomfortable with the use of a placebo control, so the placebo arm of the trial was abandoned. Detailed protocols for each treatment were developed, with medications which reflected best prescribing practice (United States Department of Health and Human Services Clinical practice guidelines 7), a treatment period of six months, and a follow-up period to track longer-term changes. The protocols were designed to be simple to use and, to minimise tasks for the GPs, telephone assessments by a clinical psychologist were planned to confirm diagnosis, assess severity, and evaluate outcome. Design: Patients assessed by their GPs as having mild to moderate major depression were to be randomly allocated to: structured problem solving alone; selective serotonin reuptake inhibitor (SSRI) medication and non-specific counselling; or SSRI medication and structured problem solving. Patients were required to meet International classification of diseases - 10th revision 8 criteria for a mild or moderate major depressive episode. Exclusion criteria included current or previous manic (or hypomanic) or psychotic symptoms, current suicidal intent, current drug or alcohol abuse, current pharmacological or psychological treatment for depression, and (to exclude severe depression) current somatic (or melancholic) features. Participating general practitioners: In 1998, the protocols were again reviewed by students in the Masters course, who agreed that they were consistent with current best primary care practice and could be delivered in this setting. Ten GPs agreed to participate - four were current second-year students in the Masters course, and six were graduates of the course. Three current second-year students did not participate because they were not working in general practice, and two because they did not wish to randomly allocate their patients to treatment. All participating doctors had been trained in assessing and managing depression (including structured problem solving), and clinical supervision was offered over the course of the trial. Back to text 2: Why were no patients recruited? Six months after commencement of the trial in 1998, despite frequent reminders and discussions about the trial protocol, no patients had been recruited. This prompted a clinical audit of all patients with depressive symptoms attending the participating doctors' practices over three weeks. Each general practitioner recorded patients presenting with depressive symptoms and the reasons why they considered them unsuitable for the trial on a form we designed for this purpose. Over the three weeks, 114 patients presented with depressive symptoms (12% of all presenting patients), but none were entered in the study. The results of the audit are shown in the Table. The most frequently cited reasons for exclusion were current pharmacological or psychological treatment for depression, or depression of insufficient severity to meet ICD-10 criteria. In an attempt to improve recruitment in the following six months, we dropped the exclusion criteria of current psychological treatment and insufficient severity (these features now to be assessed by the clinical psychologist), yet still no patients were recruited. Reasons for not entering trial No. (%) patients* Depression insufficiently severe to meet ICD-10 criteria 27 (23.6%) Severe depression or criteria met for "somatic syndrome" 13 (11.4%) Prior or current manic, hypomanic or psychotic episode 6 (5.3%) Serious suicidal intent 5 (4.4%) Current drug or alcohol abuse 6 (5.3%) Current pharmacological treatment for depression 38 (33.3%) Current psychological treatment for depression 34 (29.8%) Physical problems precluding use of an SSRI 1 (0.9%) Prior failure to respond to an SSRI 5 (4.4%) Patient refused randomisation 7 (6.1%) General practitioner not confident about using problem solving with this patient 20 (17.5%) Other (eg, dementia, communication difficulties, personality disorders) 22 (19.3%) *There were 114 patients, but general practitioners frequently nominated more than one reason for excluding patients. ICD-10=International classification of diseases - 10th revision.8 SSRI=selective serotonin reuptake inhibitor. Once the trial was formally abandoned, there were discussions with the four GPs still completing their Masters course. These GPs admitted unease with the process of randomisation, despite the demonstrated efficacy of both treatments for this population. Back to text

Caroline J Hunt · Louise M Shepherd · Gavin Andrews