Research Article REVIEW OF THE LONG-TERM EFFECTIVENESS OF COGNITIVE BEHAVIORAL THERAPY COMPARED TO MEDICATIONS IN PANIC DISORDER Deepa N. Nadiga, M.D., Paula L. Hensley, M.D., and E. H. Uhlenhuth, M.D. Panic disorder is a recurrent and disabling illness. It is believed that Cognitive Behavioral Therapy (CBT) has a long-term protective effect for this disorder. This would offer CBT considerable advantage over medication management of panic disorder, as patients often relapse when they are tapered off their medications. This is a review of the literature about the long-term effectiveness of CBT. We searched for follow-up studies of panic disorder using CBT. Of the 78 citations produced in the initial search, most bad major methodological flaws, including ignoring losses to follow-up, not accounting for interval treatment, and unclear reporting. Three papers met strict methodological criteria, and two of these demonstrated a modest protective effect of CBT in panic disorder patients. We make recommendations for well-designed studies involving comparisons of medications and cognitive behavior therapy. Depression and Anxiety 17:58-64, 2003. 2003 Wiley-Liss, Inc. Key words: adult; antidepressants; cognitive behavioral therapy; panic disorder; follow-up studies INTRODUCTION Panic disorder is a common [Weissman et al., 1997] and disabling [Klerman et al., 1991; Pollack and Marzol, 2000] illness. The disorder involves recurrent unexpected panic attacks accompanied by one of the following: a persistent concern about having additional attacks, worry about the implications of the attacks, and a significant change of behavior (DSM-IV). It is often accompanied by agoraphobia [Klerman et al., 1991]. Medications (SSRIs, TCAS, MAOIs, and benzodia- zepines) treat panic disorder effectively and achieve remission rates upwards of 70% [Ballenger, 1993; Toni et al., 2000]. However, relapse rates of 50% are common on discontinuation of antidepressant medica- tion [Toni et al., 2000]. These rates are higher with benzodiazepines [Noyes et al., 1991]. Cognitive beha- vioral therapy (CBT) is also effective in treating panic disorder with 75% of patients achieving panic-free end states [American Psychiatric Association, 1998]. In contrast, cognitive behavioral therapy is reputed to maintain its treatment gains over time, and it is in this area that CBT claims a distinct advantage over medications. Here, we assemble the evidence for this claim. The companion to this article examines the 2003 WILEY-LISS, INC. long-term effectiveness of CBT in studies of depres- sion (Hensley et al., submitted). MAJOR ISSUES IN FOLLOW-UP STUDIES DROPOUTS Acute treatment studies. Most acute treatment studies nowadays use an intent-to-treat analysis (ITT). In such an analysis, all randomly assigned subjects are included, usually using the latest available set of clinical measures (last observation carried forward; LOCF). Department of Psychiatry, School of Medicine, University of New Mexico Health Sciences Center, Albuquerque, New Mexico *Correspondence to: Dr. Deepa N. Nadiga, UNM Department of Psychiatry, 2400 Tucker N.E., 4th floor, Albuquerque, NM 87131. E-mail: dnadiga@salud.unm.edu Received for publication 14 December 2001; Accepted 8 Novem- ber 2002 Published online in Wiley InterScience (www.interscience.wiley. com). DOI 10.1002/da.10084 Subjects who drop out are considered to be treatment failures, either because of non-response or adverse events [Rickels and Schweizer, 1998]. An ITT analysis of a study prevents the "sieve effect" (i.e., patients who are benefited by treatment tend to stay on), with narrowing of the contrast between active and control conditions. Follow-up studies. Now that ITT analysis is accepted in short-term therapeutic research, the idea that dropouts have to be taken into account should be extended to follow-up studies. Currently, follow-up studies usually include only the improved subjects completing the acute treatment as a baseline. The problem is that the original randomization is lost and now we have biased subsamples. Patients who selec- tively survive an ineffective acute treatment of months' duration are likely to be hardy and at low risk of relapse after treatment is withdrawn, whereas patients who respond to an active treatment are more likely to relapse when it is withdrawn. Furthermore, patients who fail acute treatment may improve spontaneously a year or two later. Apart from ignoring attrition in the acute phase of the study, most studies also ignore attrition in the follow-up phase. The percentage of patients who maintain wellness is calculated using the total number of survivors at follow-up as a base. An assumption is made that there will be some attrition over time and these losses do not affect the results. However, this may result in over estimation of treatment effects. A better solution would be to use the last available observation. Where such data are not available, it would be more conservative to assume that losses to follow-up actually represent non-response to treatment or adverse events, as in acute treatment studies [Rickels et al., 1993]. The argument against this line of reasoning is that not all losses to follow-up represent failure of treat- ment. Subjects simply may tire of the follow-up process once gains have been made. Other losses to follow-up may occur because of subjects moving out of the geographical area. This argument certainly holds for some attrition that occurs in longer follow-up studies, but it is hard to make this case for follow-ups of two years or less. In the absence of data, all of the aforementioned arguments are speculative. The issue can be argued both ways. For the purpose of this paper we have decided to take the approach of treating the dropouts as failures of treatment. We believe that an ITT analysis using the original number randomized as the base would result in a conservative but clear outcome. CRITERIA FOR LONG-TERM OUTCOMES Many studies use cross-sectional evaluations of continuous variables to record improvement in patients with a particular intervention. This procedure fails to detect relapses that occur between evaluations. For example, a patient may have a relapse, recover completely, and present for follow-up. A cross-sec- tional assessment at this point would fail to document the interval relapse. For this purpose, full interval histories and binary definition of clinical status is required. USE OF TREATMENT IN THE FOLLOW-UP PERIOD Even in studies that are carefully carried out in the acute treatment phase, patients are free to use any treatment that they need in the follow-up period. It would be ethically inappropriate to deny such treat- ment in long-term follow-up studies. However, many of these studies do not report the number of patients who take other treatment in each experimental group. If patients require additional treatment, they should be classified as having failed the criterion of long-term effectiveness of the experimental treatment. Follow-up studies that look only at cross-sectional measurements of clinical status at different points in time without accounting for additional treatment during the inter- vals are underepresenting failures. THE HYPOTHESIS It is well known that many patients with panic disorder relapse on discontinuation of medications. The belief that even a few weeks of cognitive behavioral therapy (CBT) results in favorable long- term outcomes is widely held. We present a critical review of the evidence relevant to this belief. We identified follow-up studies that involved comparisons of CBT with medications. We wanted to examine if any of the claims of significant difference made in the original paper would remain if we performed a recalculation using ITT and preferably LOCF analysis. Using this standard, we predicted that the evidence supporting the longitudinal effectiveness of CBT in panic disorder would be modest. METHODS The initial search for relevant reports included journal articles indexed in Medline, PsychInfo, Sci- Search, SocSciSearch, and the Cochrane Library. The terms "panic disorder", "Agoraphobia", and "follow-up studies" were used. New references from reviews and primary sources were traced and added. This search produced 78 citations. The following criteria were applied to the products of the initial search: The review focused on follow-ups of straightfor- ward head-to-head comparisons of CBT and control treatments (including medications) in general samples of adult panic disorder patients. Studies of patients aged less than 18 years were excluded. Follow-up studies lacking a credible control group were excluded. Credible control groups for this purpose were taken to be patients treated with medication, who were expected to relapse after treatment was withdrawn, and patients who received a clearly inferior treatment, such as a waiting list or placebo. This requirement alone eliminated 43 studies. Studies of special samples, such as patients who failed pharmacotherapy, were excluded. Studies of exposure only (behavior treatment) were excluded, since some would argue that this is less effective and enduring than CBT. This eliminated two studies. Studies of specific elements of CBT ("decomposi- tion studies") were excluded. Follow-up studies of less than six months were excluded. Although relapse often occurs early, effects lasting less than six months can hardly be considered "long-term." This eliminated two stu- dies. There had to be a binary measure clearly defining patients who maintained remission. Histories of relapse and use of treatments during the intervals between follow-up interviews were re- quired. Only primary sources were included in this review. Eight of the originally identified citations were excluded because they were reviews, and 12 were excluded because they were on some other topic entirely. RECALCULATION Eleven studies met the previously stated criteria. For each of these studies, we determined whether appro- priate data were presented in the paper to enable a recalculation using an ITT analysis based on the original number randomized. This reduced the number of studies to three. Survivors then were identified from the follow-up group. Survivors were defined as subjects who: Are present (i.e., have not dropped out), Are doing well by standardized criteria as defined by each study, or Did not require any additional treatment in the interim. The percentage of these survivors over the original number randomized to each arm was calculated. A Fishers exact test was performed to see if the differences were significant. Paired comparisons that were significant at P TABLE 1. Sharp DM, et al. [1996] Fluvox Plac Fluvox+CBT Plac+CBT CBT Overall exact P Evaluation At randomization Measure of remission n % n % n % n % n % 36 100.0 37 100.0 38 100.0 36 100.0 43 100.0 Patients maintaining Hamilton Anxiety Scale 11 30.6 8 21.6 18 47.4 18 50.0 15 34.9 0.065 remission Symptom Rating Test 4 11.1 4 10.8 10 28.3 11 30.6 11 25.6 0.100 FQ-AG 10 27.8 7 18.9 16 42.1 17 47.2 15 34.9 0.079 At follow-up 6 months after acute treatment Effect size compared with Hamilton Anxiety Scale 0.17 0.19 0.04 Symptom Rating Test 0.19 0.19 0.14 FQ-AG 0.14 0.15 0.07 Effect size compared with Hamilton Anxiety Scale Symptom Rating Test FQ-AG 0.00 0.09 0.15 0.16 Fluvox, fluvoxamine; Plac, placebo; CBT, cognitive behavior therapy; FQ-AG, Fear Questionnaire, agarophobia subscale. Effect sizes are calculated as the difference between two portions. "Number of patients with "clinically significant." improvement from baseline 0.09 0.26 0.28 0.13 0.16 0.20 0.23 0.28 TABLE 2. Loerch B et al. [1999] + CBT+Plac CM+Moc CM+Plac Overall Exact p Evaluation Measure of remission 22 % 22 % 22 % 22 % At randomization 14 100.0 14 100.0 16 100.0 11 100.0 Patients maintaining remission Fear questionnaire 10 or less reduced 50%+ 8 57.1 2 14.3 0 0.0 0 0.0 0.000 10 71.4 3 21.4 0 0.0 0 0.0 0.000 Effect size compared with CM+Moc Fear questionnaire 10 or less reduced 50% 0.57 0.14 0.71 0.21 Effect size compared with CM+Plac Fear questionnaire 10 or less reduced 50%+ 0.57 0.14 0.00 0.71 0.21 CBT, cognitive behavior therapy; Moc, moclobemide; Plac, placebo; CM, clinical management. Effect sizes are calculated as the difference between two proportions. Paired comparisons in bold font denote difference significant at P