Early response as predictor of final remission in elderly depressed patients

Transcription

1 INTERNATIONAL JOURNAL OF GERIATRIC PSYCHIATRY Int J Geriatr Psychiatry (2009) Published online in Wiley InterScience ( Early response as predictor of final remission in elderly depressed patients Rob M. Kok 1 *, Carlijn van Baarsen 2, Willem A. Nolen 3 and Thea J. Heeren 4 1 Department of Old Age Psychiatry, Parnassia Psychiatric Institute, The Hague, The Netherlands 2 Knowledge Centre of Psychiatry in the Elderly, Zeist, The Netherlands 3 Department of Psychiatry, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands 4 Symfora Group Centers of Mental Health Care, Amersfoort, The Netherlands SUMMARY Background Several studies have attempted to predict the final response or remission based on improvement during the early course of treatment of major depression. There is however a great variation in cut offs used to define early response and in the optimal week to predict final results. Objective To compare different cut offs at different time points early in the treatment of elderly depressed patients. Method A 12 week randomised, controlled trial in 81 elderly inpatients with DSM IV major depression comparing venlafaxine with nortriptyline. At least 20, 25, 30 or 50% improvement was analysed after 1, 3 and 5 weeks using the Hamilton Depression Rating Scale and the Montgomery Asberg Depression Rating Scale. We plotted sensitivity against 1 specificity and calculated areas under the curve (AUCs). Results The highest percentage of correctly classified patients is found using at least 50% decrease as cut off in week 5, with acceptable sensitivity (81.8%) and specificity (87.4%). In week 5, the AUCs were (95% CI ) and (95% CI ) for the HAM-D and MADRS, respectively. Conclusions Combining the results from our study and the other studies addressing this issue, we suggest that the treatment should be changed in the elderly if after 3 4 weeks less than 30% improvement in depression score has been achieved. Copyright # 2009 John Wiley & Sons, Ltd. key words early improvement; final response; prediction; major depression; elderly INTRODUCTION The question when and how to decide to switch an antidepressant in depressed patients, who are not or minimally responding in the first weeks of treatment, is one of the major challenges in clinical practice. A switch too early may include patients who would have responded later in the treatment, a switch too late may unnecessarily prolong an ineffective treatment. Several studies in adult patients have attempted to predict final response or remission based on improvement during the early course of treatment as measured with *Correspondence to: R. M. Kok, Department of old age psychiatry, Parnassia Psychiatric Institute, Mangostraat 1, 2552 KS, The Hague, The Netherlands. r.kok@parnassia.nl Copyright # 2009 John Wiley & Sons, Ltd. depression rating scales. For example, Nierenberg has found that 75% of patients finally responding, defined as 50% decrease on the Hamilton Depression Rating Scale (HAM-D) at 8 weeks, had already an onset of response, defined as 30% decrease on the HAM-D, by week 4 (Nierenberg et al., 2000). Szegedi et al. (2003) found early improvement, defined as a score reduction of > 20% on the HAM-D within 2 weeks, to be a highly sensitive predictor of later stable response or stable remission. Kieser and Szegedi (2005) using 20% improvement on the HAM-D to predict a sustained 50% decrease, found a high sensitivity (87%) and low specificity (43%) in week 2 and the opposite in week 1 (43 and 86% respectively). However, another study found that patients unimproved (less than 25% improvement in the HAM-D Received 11 December 2008 Accepted 11 February 2009

2 r. m. kok ET AL. score) at week 6, had a remission rate at week 12 of 31% using an intention to treat analysis, or 41% using a completers analysis (Quitkin et al., 2003). We are aware of three studies in elderly patients in which final remission is predicted based on early response. In the first, a 20% decrease in HAM-D score at week 3 correctly classified two-third of elderly depressed patients in remission at 6 weeks (Koran et al., 1995). The second study found that only 16.5% of elderly patients with < 30% reduction in baseline HAM-D 24 score by week 4, achieved remission after 12 weeks (final HAM-D 6) (Sackeim et al., 2005). In the third study, more than two-third of elderly patients who had a partial response (HAM-D 17 score or score of 10 with a < 50% decrease from baseline), in contrast with 30% of nonresponders at week 4, became full responders (HAM-D 17 score 10 and > 50% decrease) with eight additional weeks of treatment (Mulsant et al., 2006). The variation in cut offs for defining early response (20, 25, 30, 50%) and in the best week to predict final remission (week 1, 2, 4, 6) is impressive and seems rather arbitrarily. In addition, the majority of the published studies mentioned above are either open studies (Nierenberg et al., 2000; Quitkin et al., 2003) or combined results from more than 1 RCT (Kieser and Szegedi, 2005; Sackeim et al., 2005; Mulsant et al., 2006). As a result of this, there is no gold standard to predict final remission. The aim of our study was to compare different cut offs at different time points early in the treatment of elderly depressed patients. MATERIALS AND METHODS We have recently finished a double-blind RCT of 12 weeks duration conducted in 81 depressed inpatients, aged 60 years or older. Patients meeting DSM-IV criteria for major depression and having a score of 20 on the Montgomery Asberg Depression Rating Scale (MADRS) were treated with either venlafaxine (n ¼ 40) or nortriptyline (n ¼ 41). Baseline characteristics, inclusion and exclusion criteria, dosing schedules, efficacy and tolerance data of both antidepressants have been published elsewhere (Kok et al., 2007). In short, patients had a mean age of 72.2 years (SD ¼ 7.6), were predominantly female (n ¼ 59; 72.8%), the majority had a recurrent depression (n ¼ 60; 74.1%) with nearly half of all patients having psychotic features (n ¼ 40; 49.4%). Severity of depression was evaluated at baseline and at week 1, 3, 5, 7, 9 and 12 with the HAM-D 17 and the MADRS. At baseline, the mean HAM-D score was 24.4 (SD ¼ 5.3) and the mean MADRS score was 32.9 (SD ¼ 6.2). Remission on the HAM-D 17 was defined as a score of 7, remission on the MADRS as a score of 10, both at 12 weeks. We could not demonstrate any statistically significant difference in efficacy between venlafaxine and nortriptyline and pooled both groups for the present analysis. We have calculated receiver operating characteristic (ROC) curves (a plot of sensitivity against 1 specificity using different cut offs) after 1, 3 and 5 weeks of treatment. The percentage of false negatives (patients not improving X% in week 1, 3 or 5 but achieving final remission in week 12) are of great importance because in these patients treatment may be stopped too early. To correct for dropping out, we also performed a completers analysis. RESULTS After 12 weeks of treatment, 22 out of 81 patients (27.2%) achieved remission according to the HAM-D (final score of 7) and 26 patients (32.1%) achieved remission according to the MADRS (final score of 10). The number of patients achieving at least 50% decrease in HAM-D score is nine in week 1, 23 in week 3, 25 in week 5 and 36 in week 12. For the MADRS, these numbers are four (week 1), 18 (week 3), 23 (week 5) and 38 (week 12), respectively. Figure 1 displays the ROC curves for the HAM-D 17 score after 1, 3 and 5 weeks of treatment. The ROC curves using the MADRS are almost identical (figure not displayed). Figure 1. ROC-curves using the HAM-D to predict final remission.

3 EARLY PREDICTION OF REMISSION Table 1. Prediction of final remission by early improvement using the HAM-D Prediction Sensitivity Specificity CC PPV NPV Decrease in week Decrease in week Decrease in week CC ¼ correctly classified; PPV ¼ positive predictive value; NPV ¼ negative predictive value. Using the different cut offs for defining early response on the HAM-D that were used in the studies mentioned in the Introduction (reduction of at least 20, 25, 30, 50%), the sensitivity, specificity, percentage of patients correctly classified, positive predictive value and negative predictive value in week 1, 3 and 5 are presented in percentages in Table 1. As can be calculated from Table 1, the falsenegative rates using these different cut offs varied in week 1 between 27.3 and 68.2%, in week 3 from 9.1 to 27.3% and in week 5 from 4.5 to 18.2%. The area under the ROC curves (AUC) of these weeks for the HAM-D and the MADRS is presented in Table 2. No significant difference between the HAM- D and the MADRS could be demonstrated using AUC s in predicting remission. In 14 patients the treatment was terminated according to protocol because of insufficient response and 13 patients dropped out due to serum levels outside the allowed range (n ¼ 5), side effects (n ¼ 4), Table 2. AUC using the HAM-D and the MADRS to predict final remission HAM-D MADRS AUC 95% CI AUC 95% CI Week Week Week AUC ¼ Area under the curve, CI ¼ Confidence interval. refusal (n ¼ 3) or non-compliance (n ¼ 1). Repeating the analyses with only the 54 completers did not have a significant impact on the results. DISCUSSION To the best of our knowledge, this is the first double blind RCT using ROC curves with 95% confidence intervals to compare different methods of predicting final remission based on early response. Even in week 1, prediction is significantly better than by chance (AUC ¼ 0.5). As can be expected, prediction is better according to the AUC s in later visits during the RCT, but the confidence intervals suggest that these differences may not be statistically significant. However, the limited number of included patients may be responsible for a type 2 error. The percentage of patients correctly classified as remitted or not remitted improved in later visits only when using at least 50% decrease as cut off. In contrast, the percentage of patients correctly classified decreased during later visits using at least 20 or 25% improvement to predict final remission and this is explained by the lower specificity. The highest percentage of correctly classified patients is found using at least 50% decrease as cut off in week 5, with acceptable sensitivity (81.8%) and specificity (87.4%). We are aware of only one other study in elderly patients using data from one RCT (n ¼ 671), but this study compared prediction using only one cut off ( 20% reduction) in week 1, 2 and 3 (Koran et al., 1995). Both other geriatric studies compared data from two RCTs (n ¼ 446; Sackeim et al., 2005) or three RCTs (n ¼ 472; Mulsant et al., 2006) thus improving the power but at the cost of loosing internal validity. Only one geriatric study (Mulsant et al., 2006) compared different cut offs within the same patient group and found that final remission can be predicted after 4 weeks of treatment, although this study did not present confidence intervals. Koran et al. (1995) found that their criterion to predict final remission, at least 20% reduction in week 1, 2 or 3, was not a clinically useful predictor because only 64.4% of their patients were correctly classified with regard to remission. This is in accordance with our findings in week 1, where only 54.5% of patients and in week 3 only 64.1% of patients were correctly classified using at least 20% reduction as cut-off. The third geriatric study used one cut-off, at least 30% reduction, in different weeks to predict final remission (HAM-D 6) after 12 weeks and found that patients not having at least 30% reduction in week 4 have a

4 r. m. kok ET AL. probability of only 16.5% to achieve remission (Sackeim et al., 2005). In our study, final remission could be predicted in the majority of our patients even after 1 week of treatment. However, the high percentage of false negative results suggest that after 1 week, still many patients ( %) without an early improvement finally achieve remission. According to Quitkin et al. (2003), continuation of treatment is justified if the chance of remission is at least 30%, as prior studies have suggested that this is the change of remission after switching to another antidepressant. If we apply this criterion in our patient group, patients not having at least 30% reduction in their depression score in week 3 or at least 50% in week 5, should have their treatment changed, as the chance of achieving final remission varies between 4.5 and 18.2%. Combining the results from our study and the other studies in older patients mentioned above, we suggest that the treatment should be changed in the elderly if after 3 4 weeks less than 30% improvement in depression score has been achieved. This is in accordance with a review of studies into the speed of onset of antidepressants, suggesting that if patients do not show 20% improvement within the first 2 4 weeks of treatment, their treatment regimen should be changed (Gelenberg and Chesen, 2000). Another important finding may be that the MADRS, often used in the elderly, may be as effective as the HAM-D in predicting final remission. In our opinion, one of these rating scales should be used in assessing early improvement in the treatment of every depressed elderly patient, as small changes will be difficult to detect without the use of a rating scale. Whether a selfrating scale, as for example the geriatric depression scale, is equally effective in predicting final remission has not been tested yet. The most important limitation of our study is that only 81 patients were included resulting in a lack of power finding differences between different methods to predict final remission. Another limitation in our study is that it was not a fixed dose study, as was also the case in the two more recent geriatric studies (Sackeim et al., 2005; Mulsant et al., 2006). In these studies, doses were adjusted in part based on response, which may influence the results. Only one geriatric study is a fixed dose study (Koran et al., 1995), but this has the disadvantage of differing from real practice as non-response after a few weeks of treatment often leads to dose escalation. Finally we would like to stress that our results should be considered preliminary as our analyses KEY POINTS In elderly inpatients, prediction of final remission is possible as early as week 3, Patients not having at least 30% reduction in their depression score in week 3 or at least 50% in week 5, should have their treatment changed, The HAM-D and MADRS may be equally effective in predicting the final remission. was a post-hoc analyses. Therefore, replication in larger patient samples is welcomed. DECLARATION OF INTERESTS Rob M. Kok has also received a research grant from Lundbeck and has received speaker s honoraria from GlaxoSmithKline, Lundbeck, Pfizer and Wyeth. Thea J. Heeren has received speaker s honoraria from Eli Lilly and Lundbeck. Willem A. Nolen has received research grants from Astra Zeneca, GlaxoSmithKline and Wyeth; has served as consultant for Astra Zeneca, Eli Lilly, GlaxoSmithKline, Johnson & Johnson and Pfizer and has received speaker s honoraria from Astra Zeneca, Eli Lilly, GlaxoSmithKline, Johnson & Johnson, Organon and Pfizer. Carlijn van Baarsen has no conflicts of interest. ACKNOWLEDGEMENTS This study was supported in part by a research grant from Wyeth, The Netherlands. REFERENCES Gelenberg AJ, Chesen CL How fast are antidepressants? J Clin Psychiatry 61: Kieser M, Szegedi A Predicting stable treatment response in patients with major depression treated with Hypericum extract WS 5570/5572. Pharmacopsychiatry 38: Kok RM, Nolen WA, Heeren TJ Venlafaxine versus nortriptyline in the treatment of elderly depressed inpatients. A randomised, double-blind, controlled trial. Int J Geriatr Psychiatry 22: Koran LM, Hamilton SH, Hertzman M, Meyers BS, Halaris AS, Tollefson GD, et al Predicting response to fluoxetine in geriatric patients with major depression. J Clin Psychopharmacol 15: Mulsant BH, Houck PR, Gildengers AG, Andreescu C, Dew MA, Pollock BG, et al What is the optimal duration of a short-

5 EARLY PREDICTION OF REMISSION term antidepressant trail when treating geriatric depression? J Clin Psychopharmacol 26: Nierenberg AA, Farabaugh AH, Alpert JE, Gordon J, Worthington JJ, Rosenbaum J, et al Timing of onset of antidepressant response with fluoxetine treatment. Am J Psychiatry 157: Quitkin FM, Petkova E, McGrath PJ, Taylor B, Beasley C, Steward J, et al When should a trial of fluoxetine for major depression be declared failed? Am J Psychiatry 160: Sackeim HA, Roose SP, Burt T Optimal length of antidepressant trails in late-life depression. J Clin Psychopharmacol 25 (suppl 1): S34 S37. Szegedi A, Müller M, Anghelescu I, Klawe C, Kohnen R, Benckert O Early improvement under mirtazapine and paroxetine predicts later stable response and remission with high sensitivity in patients with major depression. J Clin Psychiatry 64: