The purpose of this study was to examine whether experimental and observational pharmacoeconomic analyses of antidepressant drugs support the choice of one of the selective serotonin reuptake inhibitors or newer antidepressants as first-line treatment for patients with major depression. We systematically reviewed economic evaluations of two or more antidepressants completed in clinical practice. A systematic electronic search yielded 38 studies meeting the inclusion criteria, of which 23 were administrative database analyses, 12 were observational studies, and 3 were randomized clinical trials. Experimental data indicated that tricyclic antidepressants are equivalent to selective serotonin reuptake inhibitors in terms of total expenditure. While the database analyses are susceptible to bias and confounding variables, they provided an added dimension based on observations from everyday clinical practice. The majority of these studies failed to show any significant difference. Taken together, available pharmacoeconomic studies indicate that tricyclic drugs and selective serotonin reuptake inhibitors have similar cost effectiveness in the health care systems where these comparisons have been made.
In the past 10–15 years, antidepressant prescription patterns have dramatically changed, mainly reflecting an increase in the use of the selective serotonin reuptake inhibitors (SSRIs). This dramatic change in antidepressant use has produced a progressive increase in total drug expenditures for antidepressants. 1,2
Although many clinical recommendations and prescribing guidelines have summarized the existing evidence on antidepressant efficacy and tolerability, 3–7 these reports have only marginally taken into account pharmacoeconomic data. This might reflect the low number of clinical trials considering costs associated with antidepressant treatment. However, there are many studies addressing these issues with observational designs, including retrospective administrative database analyses, studies based on pharmacoeconomic models, and prospective observational studies. The present systematic review aims at investigating whether experimental and observational pharmacoeconomic analyses of antidepressant drugs support the choice of one of the SSRIs or newer antidepressant drugs as first-line treatment of depression.
*Department of Medicine and Public Health, Section of Psychiatry, University of Verona, Italy; †Department of Psychiatry, Hospital of Legnano, Milan, Italy; and ‡Division of Psychological Medicine, Guy’s, King’s, and St. Thomas’ School of Medicine and Institute of Psychiatry, London, UK
Received February 15 2002; accepted June 27, 2002.
Address requests for reprints to: Dr. Corrado Barbui, Department of Medicine and Public Health, Section of Psychiatry, University of Verona, Ospedale Policlinico, 37134 Verona, Italy. Address e-mail to: firstname.lastname@example.org.
Only economic evaluations of two or more antidepressant drugs conducted in clinical practice were included. Specifically, decision-analytic models based on assumptions derived from previous studies were excluded, while clinical trials, observational studies, and database analyses were included.
Relevant studies were located by adopting a two-step procedure. First, the Cochrane Collaboration Depression, Anxiety, and Neurosis Controlled Trials Register (CCDANCTR) was searched with no year limit for the following terms: depression and paroxetine or sertraline or fluoxetine or fluvoxamine or citalopram or venlafaxine or reboxetine or mirtazapine or nefazodone or trazodone or moclobemide or amitriptyline or nortriptyline or clomipramine or imipramine or desipramine or dothiepin. Second, observational and database studies were located through a specific MEDLINE and EMBASE electronic search performed with no year limits. The subject headings cost and cost analysis and antidepressive agents were exploded. Reference lists of relevant papers and previous reviews were hand-searched for published reports missed by the electronic search. No data were obtained through direct contact with the pharmaceutical industry.
A standard form was used by two reviewers who independently extracted information on the year of publication, study design, length of follow-up, country and setting of the study, antidepressants compared, sample size, study inclusion criteria, patient age, and diagnosis. Additional information on study results was extracted according to the following three outcome measures: antidepressant drug costs, depression-related costs, and total health care costs.
Included reports were grouped, according to the study design, as follows. Database analyses were retrospective economic studies carried out with use of administrative medical claims databases. Observational studies were those with a prospective, retrospective, or crosssectional design, carried out with the specific aim of comparing the economic outcome of different antidepressant therapies. Randomized clinical trials (RCTs) were studies in which patients were randomly allocated to competitive antidepressant treatments.
Study results were considered with use of the following outcome measures. Antidepressant prescription costs were all antidepressant pharmaceutical costs during the study period. Depression-related costs were antidepressant prescription costs (a) plus outpatient, inpatient, and laboratory costs associated with a depression diagnosis code. Total health care costs were depression-related costs (b) plus all outpatient, inpatient, laboratory, and pharmaceutical costs, irrespective of diagnosis. Other costs were costs that could not be placed into one of the above categories.
Since most studies did not include effectiveness data, the cost-effectiveness plane or a permutation matrix showing possible outcomes in terms of costs and effectiveness was not feasible. 8,9 Therefore, a descriptive summary was used to summarize study findings, in conjunction with a tabular approach to recording the results. 10
The electronic search yielded 38 studies meeting the inclusion criteria, 11–48 of which 23 were administrative database analyses, 12 were observational studies, and 3 were randomized clinical trials.
All but one of the 23 administrative database analyses were carried out in the United States, the exception being 1 study conducted in the United Kingdom. 47 All these analyses adopted a similar methodology, based on the identification of patients who had a paid insurance claim that indicated a diagnosis of depression and treatment with antidepressants (Table 1). In the majority of cases, patients were assigned to a drug cohort on the basis of their initial prescription, and resource utilization data 6–24 months after the first prescription were then analyzed.
Observational studies included two prospective, eight retrospective, and two cross-sectional studies (Table 1). The three randomized clinical trials included a comparison of sertraline and fluoxetine carried out in general practice over a 6-month period in France, 13 a clinical trial with 6 months of follow-up carried out in the Czech Republic, 24 and a pragmatic randomized trial comparing fluoxetine, imipramine, and desipramine in primary care in the United States (Table 1). 39
Findings from database analyses
Fifteen studies provided information on antidepressant drug acquisition costs. In 11 studies, antidepressant costs were higher for the SSRIs and newer drugs in comparison with tricyclic drugs (TCAs;Table 2), whereas four of five head-to-head comparisons of SSRIs found fluoxetine to be more expensive than other SSRIs (Table 2). Seven database analyses provided information on depressionrelated costs, with contrasting findings (Table 2). Fifteen database analyses calculated total health care costs associated with SSRI or TCA treatment: in eight cases differences were detected, always favoring the SSRIs (Table 2). Twenty of 23 database analyses were financially supported by pharmaceutical industries or conducted by employees of the pharmaceutical industry.
Findings from quasi-experimental studies
Three studies showed that antidepressant drug acquisition costs were higher for the SSRIs and newer drugs in comparison with TCAs. Data on depression-related cost, available in only one case, showed that tricyclic drugs were more costly than sertraline (Table 3). Three studies calculated total health care costs, with contrasting results. Five of 12 observational studies calculated the cost consequences of antidepressant overdose; these analyses found that hospital and medical costs attributable to drug overdose were higher for patients who overdosed on tricyclic drugs than for those who overdosed on SSRIs (Table 3). Finally, two reports showed that the risk of work loss was lower for patients taking fluoxetine rather than other treatments. Seven reports made a declaration of competing interests.
Findings from randomized clinical trials
Fluoxetine, in comparison with imipramine and desipramine in the study carried out by Simon and colleagues, was associated with higher antidepressant drug acquisition costs (Table 4). Fluoxetine and citalopram, in addition, were associated with higher antidepressant drug acquisition costs in comparison with amitriptyline in the study carried out by Hòsak and colleagues. Simon and colleagues failed to find significant differences between fluoxetine and tricyclic drugs for total health care costs; similar results were obtained by Hòsak and colleagues, while Boyer and colleagues showed that the risk of work loss was lower for patients taking sertraline than for fluoxetine recipients (Table 4).
Overall total health care costs
Table 5 presents the number of studies showing or failing to show a difference in total health care costs between TCAs and SSRIs. The majority of studies failed to show any significant difference.
The majority of studies did not include effectiveness data, and therefore a cost-effectiveness analysis could not be performed. This is a major limitation because antidepressant economic evaluations require safety and effectiveness data generated from the same samples of patients who were the object of the economic evaluation. This design provides a comprehensive assessment of new medicines and allows any economic advantages to be balanced against possible disadvantages in terms of efficacy and safety. 49
In this review the lack of clinical data meant that only economic information was extracted. This economic information, moreover, could not be summarized with the adoption of a quantitative approach. This represents a second study limitation because the size of any differences in costs between competing antidepressants was not estimated. Even for studies that did not find a difference, this information would have been relevant to distinguish between studies with a low statistical power and studies where there were really no differences. Unfortunately, heterogeneity among included studies in terms of data analysis and presentation did not allow pooling of findings.
Some studies analyzed actual or log-transformed costs, and others performed rank transformations; in a minority of cases, several analytical models were presented. In most instances crude numbers were not reported. Multivariate analyses were frequently conducted to adjust for confounding variables, and estimates of factors affecting expenditures, expressed in monetary terms against a reference category, were usually reported. Unfortunately, these estimates were not suitable for reanalysis. Even resource utilization figures were rarely reported in a way suitable for reanalysis. Therefore, a tabular approach to recording the results was used to summarize each study’s findings. 10
The nonrandomized study design of many of these papers is another major limitation. In these studies there is the possibility of confounding or selection bias. A patient’s medical and social history and the physician’s characteristics may influence both the initial choice of antidepressant and the subsequent outcome. For example, it is possible that clinicians are more likely to prescribe TCAs for individuals with more severe depression or for those who had a previous episode for which TCAs were prescribed.
Furthermore, patients who are able to work may be prescribed SSRIs because these drugs are perceived as less disruptive and more tolerable. Database analyses employed a wide range of statistical methods to control for both observed factors as well as unobserved factors correlated with initial treatment selection and subsequent outcomes. It is impossible to determine whether these techniques really controlled for confounding variables.
Study reports presented many different analytical approaches, and sometimes more than one approach was followed: including and excluding explanatory variables, using log or rank transformations or actual costs, and performing parametric and nonparametric statistical analysis. Unfortunately, different and sometimes contradictory data emerged. This leaves uncertainty about study findings and also raises concern about the analyses themselves. We note with concern that the majority of these studies were funded by individual pharmaceutical companies, and there are clearly competing interests.
This raises the strong possibility of publication bias. Although differences were detected between individual SSRIs, no company published data that indicated their compound was less cost-effective than a rival. Furthermore, the complex statistical analyses described in many of these studies leave them open to a degree of subjectivity. With many variables to enter into multivariate models, it is possible that the final results chosen tended to exaggerate differences between compounds that were advantageous to the sponsoring company.
One of the major claims supporting SSRIs over TCAs comes from the safer tolerability profile these compounds have when taken in overdose. Deliberate self-harm is a relatively rare event, and even large database analyses may not capture episodes. However, the studies that assessed the costs arising from hospitalization following TCA overdoses as opposed to SSRI overdoses indicated that there are important cost differences, with TCA overdoses being significantly more expensive. 21,26,31,32,43
Although a limitation of these studies is the possibility of confounding variables or selection bias, one advantage is the possibility of generating information useful in everyday clinical practice. In everyday clinical practice an important issue is the proportion of patients who discontinue treatment. Experimental studies estimated that around one-third of subjects drop out after 4–8 weeks of therapy, 50 and in everyday conditions this proportion might be even greater. 51 Costs are associated with these patients, who might receive other pharmacological and/or nonpharmacological treatments and might be frequent users of psychiatric and/ or medical services. For example, the risk of withdrawal symptoms at treatment discontinuation is not equivalent among antidepressants. Data from the World Health Organization’s adverse effects monitoring system, which collects spontaneous case reports from 47 countries, showed that the rate of spontaneous reporting of symptoms associated with withdrawal was higher for paroxetine than for the other antidepressant drugs. 52 This may be associated with differences in long-term costs. Unfortunately, most database analyses included only patients with continuous enrollment in the insurance plan, thus systematically excluding dropouts.
The best evidence comes from RCTs, which seem to indicate that TCAs are equivalent to SSRIs in terms of total expenditure. While the database analyses are susceptible to bias and confounding, they provide an added dimension in that they are based on observations from “real world” practice. Clinicians, managers, consumers, and researchers need high-quality clinical databases to collect information on large numbers of typical patients followed under routine circumstances. This information serves to aid clinical practice and evaluation of health technologies. 53 However, we have seen that there are too many potential threats to the validity of these reports to draw a firm conclusion that SSRIs are a cheaper option overall. We suspect that TCAs and SSRIs have approximately similar cost-effectiveness in the health care systems where these comparisons have been made.
However, there are problems of generalizability. Not only do prescribing costs differ dramatically between settings, but also health care systems are organized in very different ways. In some countries, for example, more expensive interventions (such as hospitalization) might occur routinely in cases in which treatment has failed. In others, scarce resources would make this unlikely, except in severe cases.
Therefore, the issue of generalizability in economic analyses is much more important than in standard RCT, where clinical responses to treatments are likely to be less sensitive to the setting. Moreover, within the past year in the United States, fluoxetine lost patent protection, and generic fluoxetine is now considerably less expensive than before. Similarly, other patents are expiring, and this will presumably lead to steeper reductions in costs. It is therefore difficult to predict whether available data will apply to the post-patent-protection context. As health care providers in different settings are those who ultimately pay for new innovations, it seems appropriate that they commission research on cost-effectiveness.
The authors thank Hugh McGuire, CCDANCRT Trial Search Coordinator, for assisting in developing the search strategy of this research, and Dr. Ladislav Hosák, for allowing the inclusion of information from his clinical trial.
1. Eccles M, Freemantle N, Mason J. North of England evidence-based guideline development project: summary version of guidelines for the choice of antidepressants for depression in primary care. Fam Practice 1999; 16:103–11.
2. Barbui C, Campomori A, Mezzalira L, et al. Psychotropic drug use in Italy 1984–1999: impact of regulatory changes. Int Clin Psychopharmacol 2001; 16:227–33.
3. Hotopf MH, Lewis G, Normand C. Are SSRI a cost effective alternative to tricyclics? Br J Psychiatry 1996; 168:404–409.
4. Schulberg HC, Katon W, Simon G, et al. Treating major depression in primary care practice. An update of the Agency for Health Care Policy and Research Practice Guidelines. Arch Gen Psychiatry 1998; 55:1121–7.
5. Crismond ML, Trivedi M, Pigott TA, et al. The Texas Medication Algorithm Project: report of the Texas consensus conference panel on medication treatment of major depressive disorder. J Clin Psychiatry 1999; 60:142–56.
6. American Psychiatric Association. Practice guideline for the treatment of patients with major depressive disorder (revision). Am J Psychiatry 2000; 157(suppl 4):1S–45S
7. Canadian Psychiatric Association. Clinical guidelines for the treatment of depressive disorders. Can J Psychiatry 2001; 46(suppl 1):38S–58S.
8. Black WC. The CE plane: a graphic representation of cost-effectiveness. Med Decis Making 1990; 10:212–14.
9. Geisler E, Heller O, eds. Managing Technology in Healthcare. Norwell, MA: Kluwer Academic, 1996.
10. Nixon J, Khan KS, Kleijnen J. Summarising economic evaluations in systematic reviews: a new approach. BMJ 2001; 322:1596–8.
11. Berndt ER, Russell JM, Miceli R, et al. Comparing SSRI treatment costs for depression using retrospective claims data: the role of nonrandom selection and skewed data. Value Health 2000; 3:208–21.
12. Beuzen JN, Ravily VF, Souetre EJ, et al. Impact of fluoxetine on work loss in depression. Int Clin Psychopharmacol 1993; 8:319–21.
13. Boyer P, Danion JM, Bisserbe JC, et al. Clinical and economic comparison of sertraline and fluoxetine in the treatment of depression. A 6-month double-blind study in a primary care setting in France. Pharmacoeconomics 1998; 13:157–69.
14. Claxton AJ, Chawla AJ, Kennedy S. Absenteeism among employees treated for depression. J Occup Environ Med 1999; 41:605–11.
15. Croghan TW, Lair TJ, Engelhart L, et al. Effect of antidepressant therapy on health care utilization and costs in primary care. Psychiatr Serv 1997; 48:1420–6.
16. Croghan TW, Obenchain RL, Crown WE. What does treatment of depression really cost? Health Affairs 1998a; 17:198–208.
17. Croghan TW, Melfi CA, Crown WE, et al. Cost-effectiveness of antidepressant medications. J Ment Health Policy Econ 1998b; 1:109–17.
18. Croghan TW, Kniesner TJ, Melfi CA, et al. Effect of antidepressant choice on the incidence and economic intensity of hospitalisation among depressed individuals. Adm Policy Ment Health 2000; 27: 183–95.
19. Crown WH, Hylan TR, Meneades L. Antidepressant selection and use and healthcare expenditures. An empirical approach. Pharmacoeconomics 1998; 13:435–48.
20. Crown WH, Treglia M, Meneades L, et al. Long-term costs of treatment for depression: impact of drug selection and guideline adherence. Value Health 2001; 4:295–307.
21. D’Mello DA, Finkbeiner DS, Kocher KN. The cost of antidepressant overdose. Gen Hosp Psychiatry 1995; 17:454–55.
22. Forder J, Kavanagh S, Fenyo A. A comparison of the cost-effectiveness of sertraline versus tricyclic antidepressants in primary care. J Affect Disord 1996; 38:97–111.
23. Griffiths RI, Sullivan EM, Frank RG, et al. Medical resource use and cost of venlafaxine or tricyclic antidepressant therapy following selective serotonin reuptake inhibitor therapy for depression. Pharmacoeconomics 1999; 15:495–505.
24. Hosak L, Tuma I, Hanus H, et al. Costs and outcomes of use of amitriptyline, citalopram and fluoxetine in major depression: exploratory study. Acta Medica (Hradec Kralove) 2000; 43:133–7.
25. Hylan TR, Crown WH, Meneades L, et al. Tricyclic antidepressant and selective serotonin reuptake inhibitors antidepressant selection and health care costs in the naturalistic setting: a multivariate analysis. J Affect Disord 1998; 47:71–9.
26. Kapur N, House A, Creed F, et al. Cost of antidepressant overdose: a preliminary study. Br J Gen Practice 1999; 49:733–4.
27. McCombs JS, Nichol MB, Stimmel GL. The role of SSRI antidepressants for treating depressed patients in the California Medicaid (Medi-Cal) program. Value Health 1999; 2:269–80.
28. Melton ST, Kirkwood CK, Farrar TW, et al. Economic evaluation of paroxetine and imipramine in depressed outpatients. Psychopharmacol Bull 1997; 33:93–100.
29. Nurnberg HG, Hensley PL, Thompson PM, et al. Modelling the pharmacoeconomic cost of three selective serotonin reuptake inhibitors. Psychiatr Serv 1999; 50:1351–3.
30. Ozminkowski RJ, Hylan TR, Melfi CA, et al. Economic consequences of selective serotonin reuptake inhibitor use with drugs also metabolised by the cytochrome P-450 system. Clin Ther 1998; 20:780–96.
31. Ramchandani P, Murray B, Hawton K, et al. Deliberate self poisoning with antidepressant drugs: a comparison of the relative hospital costs of cases of overdose of tricyclic with those of selective-serotonin re-uptake inhibitors. J Affect Disord 2000; 60:97–100.
32. Revicki DA, Palmer CS, Phillips SD, et al. Acute medical costs of fluoxetine versus tricyclic antidepressants. A prospective multicentre study of antidepressant drug overdoses. Pharmacoeconomics 1997; 11:48–55.
33. Sacristan JA, Gilaberte I, Hylan TR, et al. Costes del tratamiento con nuevos antidepresivos en la pratica clinica. Atencion Primaria 1999; 23:15–16.
34. Sclar DA, Robison LM, Skaer TL, et al. Antidepressant pharmacotherapy: economic outcomes in a health maintenance organisation. Clin Ther 1994; 16:715–29.
35. Sclar DA, Robison LM, Skaer TL, et al. Antidepressant pharmacotherapy: economic evaluation of fluoxetine, paroxetine and sertraline in a health maintenance organisation. J Int Med Res 1995; 23:395–412.
36. Sclar DA, Skaer TL, Robison LM, et al. Economic outcomes with antidepressant pharmacotherapy: a retrospective intent-to-treat analysis. J Clin Psychiatry 1998; 59(suppl 2):13–17.
37. Sclar DA, Skaer TL, Robison LM, et al. Economic appraisal of citalopram in the management of single-episode depression. J Clin Psychopharmacol 1999; 19(suppl 1):47S–54S.
38. Simon GE, Fishman P. Cost implications of initial antidepressant selection in primary care. Pharmacoeconomics 1998; 13:61–70.
39. Simon GE, Heiligenstein JH, Revicki DA, et al. Long-term outcomes of initial antidepressant drug choice in a “real world” randomised trial. Arch Fam Med 1999; 8:319–25.
40. Skaer TL, Sclar DA, Robison LM, et al. Economic valuation of amitriptyline, desipramine, nortriptyline, and sertraline in the management of patients with depression. Curr Ther Research 1995; 56:556–67.
41. Skaer TL, Sclar DA, Robison LM, et al. Antidepressant pharmacotherapy: effect on women’s resource utilisation within a health maintenaince organisation. J Appl Ther 1996; 1:45–52.
42. Smith W, Sherrill A. A pharmacoeconomic study of the management of major depression: patients in a TennCare HMO. Med Interface 1996; 9:88–92.
43. Stoner SC, Marken PA, Watson WA, et al. Antidepressant overdose and resultant emergency department services: the impact of SSRIs. Psychopharmacol Bull 1997; 33:667–70.
44. Sullivan EM, Griffiths RI, Frank RG, et al. One-year costs of second-line therapies for depression. J Clin Psychiatry 2000; 61:290–8.
45. Tarricone R, Fattore G, Gerzeli S, et al. The costs of pharmacological treatment for major depression. The Italian Prospective Multicentre Observational Incidence-Based study. Pharmacoeconomics 2000; 17:167–74.
46. Tollefson GD, Souetre E, Thomander L, et al. Comorbid anxious signs and symptoms in major depression: impact on functional work capacity and comparative treatment outcomes. Int Clin Psychopharmacol 1993; 8:281–93.
47. Treglia M, Neslusan C, Dunn RL. Fluoxetine and dothiepin therapy in primary care and health resource utilisation:evidence from the United Kingdom. Int J Psychiatry Clin Practice 1999; 3:23–30.
48. Viale GL. An economic analysis of physicians’ prescribing of selective serotonin reuptake inhibitors. Hospital Pharmacy 1998; 33:847–50.
49. Altamura AC, Percudani M. The use of antidepressants for long-term treatment of recurrent depression: rationale, current methodologies, and future directions. J Clin Psychiatry 1993; 54(suppl):29–37.
50. Barbui C, Hotopf M. Amitriptyline versus the rest: still the leading antidepressant after 40 years of randomised controlled trials. Br J Psychiatry 2001; 178:129–44.
51. Percudani M, Belloni G, Contini A, et al. Monitoring community psychiatric services in Italy: differences between patients who leave care and those who stay in treatment. Br J Psychiatry 2002; 180:254–9.
52. Stahl MMS, Lindquist M, Petterson M, et al. Withdrawal reactions with selective serotonin re-uptake inhibitors as reported to the WHO system. Eur J Clin Pharmacol 1997; 53:163–9.
53. Black N. High quality clinical databases: breaking down barriers. Lancet 1999; 353:1205–6.