- Email: [email protected]
or post-traumatic stress disorder
Personalized medicine in psychiatry xxx (2017) xxx–xxx
Contents lists available at ScienceDirect
Personalized medicine in psychiatry journal homepage: www.elsevier.com/locate/pmip
Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder Farrokh Alemi a,⇑, Aryan Mazloum-Yazdi a, Lora Peppard b a b
George Mason University, Department of Health Administration and Policy, United States George Mason University, School of Nursing, United States
a r t i c l e
i n f o
Article history: Received 22 May 2017 Accepted 15 July 2017 Available online xxxx
a b s t r a c t Importance: Citalopram is a common antidepressant widely used and STAR*D study has indicated that most, more than 60%, of patients who receive it, do not benefit from their first antidepressant. Objective: To identify whether citalopram is effective for patients with particular medical history. Design: We used stratification to balance case/control study design. Setting: We used the National Institute of Mental Health’s STAR*D data, collected from 41 specialty and primary medical care settings spanning across a seven year time period. Participants: The data included .mental and physical diagnoses of 4041 patients with major depression. Intervention(s) (for clinical trials) or exposure(s) (for observational studies): Patients comorbidities were balanced using Stratified Covariate Balancing method. This algorithm uses Markov Blanket of both treatment and outcome to identify which comorbidities can be ignored. It uses stratification to remove confounding introduced by the covariates in the Markov Blanket of treatment and outcome. Main outcome(s) and measure(s): A patient’s remission was judged by weather there was a 50% reduction in the Hamilton Rating Scale for Depression. Results: Patients exposed to neurological disorders were less likely to experience remission than controls without neurological disorder (odds = 0.74, chi-square=, alpha < 0.05). For exposure to PTSD, we stratified psychiatric illness in the Markov Blanket of PTSD and Neurological disorders in the Markov Blanket of Remission. The un-confounded odds of response for patients exposed to PTSD was 0.53 (Chi-square = 8.31, alpha < 0.05.) Conclusions and relevance: Patients exposed to PTSD or neurological disorders were less likely to experience remission from depression. These data suggest that citalopram should not be a first line agent prescribed for patients with these disorders. Ó 2017 Elsevier Inc. All rights reserved.
Background More than 60% of depressed patients do not benefit from their first antidepressant [1–4], prompting some experts to argue that ‘‘the bar for antidepressants has been set far too low” [5]. To understand the magnitude of the problem, consider that of the 11% of Americans currently taking antidepressants [6,7], 25 million initially took an antidepressant that was not therapeutic. Months later, after repeated trials on antidepressants, more than 50% of the initial non-respondentseventually benefited from a different antidepressant [8]; indicating that these patients are not treatment resistant and more careful selection of an antidepressant could
⇑ Corresponding author at: 4400 University Drive, Fairfax, VA 22030, United States. E-mail addresses: [email protected] (F. Alemi), [email protected] (A. Mazloum-Yazdi).
provide an earlier response. During the months of mistrial of antidepressants, patients continued to suffer from debilitating symptoms, many remained at risk of suicide and millions of dollars were spend on unnecessary medication [9]. Unfortunately, clinicians have limited guidance on which antidepressant might be the most effective for patients with different medical and psychiatric histories. Guidelines and previous studies have focused on the effectiveness of citalopram and not on characteristics of those who may or may not respond to this medication [10,11]. This paper provides clinicians with additional guidance on who would most likely respond to citalopram, a commonly prescribed antidepressant and one of the first selective serotonin re-uptake inhibitors. The task of choosing an initial antidepressant is complex. Guidelines currently recommend basing the decision on the anticipated side effects, the safety or tolerability of these side effects for the individual patient, pharmacological properties of the medication, medication response in prior episodes, cost, and patient pref-
http://dx.doi.org/10.1016/j.pmip.2017.07.003 2468-1717/Ó 2017 Elsevier Inc. All rights reserved.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
2
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
erence [10]. While a great deal of information is available in numerous randomized clinical studies [11,12], these studies have excluded patients with major comorbidities, making it difficult to anticipate response to treatment for patients with comorbidities. A patient’s race, gender, employment, education, and income have been shown to affect response to treatment [13]. Length of depression episode [14], cognition [22], and the presence of the following have also been shown to affect response to antidepressants: concurrent anxiety [15], substance use disorders [16,17], obesity [18], insomnia [19], cerebrovascular diseases [20,21], hormone imbalances [22–24], and post-traumatic stress disorder [25]. Many of these predictors do not have a consistent impact on response to treatment, making it more difficult for clinicians to decide among various treatment alternatives [26]. In addition, there are more than 24 types of antidepressants currently on the market [27], and many of them have several different brand names and can be prescribed in different dosages, further complicating the choice of the first antidepressant. Attempts to improve the precision of antidepressant selection have had mixed results. Some studies have tried to use genetic markers to predict response with mixed success. One study found these markers work for patients with psychiatric disorders [28,29]. Another study found that genetic markers could accurately (78% correct) predict response in 2/3rd of cases but not all cases [30]. Others have tried to improve the precision of antidepressant selection by dividing depression into sub-types. One study identified 5 different symptom profiles predictive of response to antidepressants [31]. Unfortunately, these profiles did not map to patients’ diagnoses (with the exception of melancholic depression), raising the possibility that these profiles were statistically significant but not clinically useful. In this paper, we examine if diagnoses from the Diagnostic and Statistical Manual of Mental Disorders can be used to predict response to antidepressants. Such an approach is advantageous because clinicians can readily implement the study findings into practice. Methods Study sample We used the STAR*D database, the largest available data set capturing response to antidepressants. These data include genetic markers and phenotypes of 4,041 patients with major depression. Demographic data (gender, age, race), data from 13 disorders (cardiovascular, vascular, hematopoietic, eyes, gastrointestinal, renal, genitourinary, musculoskeletal, neurological, psychiatric, respiratory, liver, and endocrine diagnostic disorders), and verified depression comorbidities (Alcohol, Cannabis, Panic, Posttraumatic Stress, and Generalized Anxiety disorders) were examined in this study. Data were collected from 41 clinical sites around the country, which included both specialty care settings and primary medical care settings, over a seven year time period. The response to citalopram was measured using the Hamilton Rating Scale for Depression (HRSD). Patient remission was recognized as a 50% decrease in HRSD score from the start of the trial. Method of analysis In order to accurately predict the response to citalopram for patients having a particular diagnosis, the influence of patient comorbidities should be accounted for. Until recently statistical approaches to control for comorbidities were not available. In 1983, Rosenbaum and Rubin proposed the use of propensity scoring to balance rates of occurrence of covariates/comorbidities [32]. Since then, the approach has been revised and used widely [33–
46]. We used Stratified Covariate Balancing, where the propensity weights are derived analytically without statistical modeling and through stratification [47]. All patients had received citalopram and were assigned to cases and controls based on exposure to specific disorders. Thus, in evaluating the impact of exposure to post-traumatic stress disorder (PTSD) on remission, we considered all patients with PTSD as cases, all patients without PTSD as controls, and all other comorbidities as covariates that should be balanced through stratification. One limitation of using stratification to control for effect of comorbidities is that as the number of covariates increases, only a few cases fall within each stratum. A number of methods are available to restrict the stratification to a smaller number of covariates. One approach that has been shown to be theoretically sound is to restrict the analysis to the Markov Blanket of outcome [48,49]. We fit a Bayesian probability network to the data to identify the covariates that are in the Markov Blanket of remission and them identified a temporal sequence between any two diagnoses using age at which the patients reported the diagnosis to accurately fit the network. The use of temporal sequence among a pair of variables to improve accuracy of network modeling has been reported elsewhere [50]. Diagnoses that occur later were prohibited from affecting earlier events. The BNLearn software package in R version (3.31) was used to fit the network model. Once the Markov Blanket for both (a) the exposure variable and (b) remission variables were identified, these covariates were stratified and the un-confounded impact of exposure on remission reported.
Results Table 1 shows the demographics and comorbidities of patients in our sample. To show how patients exposed to specific disorders have different comorbidities and demographics, Table 1 reports the data for exposure to PTSD and neurological disorders. Patients exposed to PTSD differed from patients not exposed to PTSD in nineteen covariates. Patients exposed to neurological disorders also differed from patients not exposed to neurological disorders in twenty covariates. These data show significant variations in demographics and comorbidities. The impact of exposure to either PTSD or to neurological disorders on response to citalopram cannot be estimated without statistically controlling for the listed comorbidities. We calculated the average age at which each diagnoses occurs to understand what variables are in the Markov Blanket of exposure and outcome. Age, race, and gender are events that occur at birth. The outcome variable occurs at end of the data collection effort and therefore no disorders can occur after it. Fig. 1 provides the age at which various diagnoses are given. For example, at alpha levels less than 0.05, cannabis abuse (Average Age = 411.30 months, Standard Deviation = 130.73 months) occurs prior to heart disease (Average = 589.76 months, standard deviation = 157.49 months). Based on these data, we organized the diagnoses into 8 categories as follows: Time 1: Age, Gender, Race Time 2: Panic Disorder, Cannabis Abuse Time 3: Neurological Disorder, General Anxiety Disorder, Psychiatric Illness, Alcohol Abuse Time 4: Gastrointestinal Disorder, Respiratory Disease, Hematopoietic Disorder, Eyes/Ears/Nose/Throat/Larynx Disorder Time 5: Musculoskeletal/Integument Disorder, Genitourinary Disorder, Renal Disease Time 6: Heart Disease, Endocrine Disorder, Liver Disease, PostTraumatic Stress Disorder Time 7: Vascular Disease Time 8: Response to Citalopram
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
3
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx Table 1 Rate of Various Events in STAR*D Data before Covariate Balancing. Neurological Disorder Cases Age 41+ Age 18–41 Gender Female Gender Male Race White Race Not-White Comorbidities
No comorbidities Heart Vascular Hematopoietic Eyes Gastrointestinal Renal Genitourinary Musculoskeletal Psychiatric Illness Respiratory Liver Endocrine Alcohol Abuse Cannabis Panic Generalized Anxiety Total
395 345 491* 249* 557 183 24** 145** 218** 81* 442** 397** 65** 268** 417** 171** 319** 102** 245** 43 12 38 51 740
Controls without Neurological Disorder 1102 1061 1299* 864* 1726 437 300** 595** 482** 160* 841** 694** 89** 417** 850** 389** 595** 149** 452** 110 31 86 114 2163
PTSD Cases **
115 43** 57** 101** 100** 58** 1** 40** 62** 15 98** 70 12 59** 110** 136** 70** 16 46 13 3 18** 13 158
Controls without PTSD 1382** 1363** 1733** 1012** 2183** 562** 324** 386** 638** 226 1185** 1021 142 626** 1157** 424** 844** 235 651 140 40 106** 152 2745
* Statistically significant at 0.05, ** Statistically significant at 0.01.
Fig. 1. Age of Occurrence of Various Diseases. Cell entries indicate > if row variables occurs after and < if it occurs before the column variable. Blank entries indicate pairs of diseases with no significant difference in age of occurrence.
In fitting a network model to the data, we prohibited the algorithm from allowing a later event to influence an earlier event. The resulting network structure is provided in Fig. 2. Certain variables (specific phobia, obsessive compulsive disorder, other disorders, and ethnicity) were not connected to any member of the network and were dropped from the analysis. The network shows the Markov Blanket of response to citalopram. The Markov Blanket is the set of variables that directionally separate the network from response to citalopram. This includes two variables: Neurological disorder and PTSD. Furthermore, the Markov Blanket of exposure to PTSD includes Psychiatric Illness. To find the effect of PTSD on
response after taking citalopram, we need to stratify the Markov Blanket of PTSD as well as the Markov Blanket of response; in short we need to only stratify two (Neurological Disorder and PTSD) of the twenty-one possible covariates. Fig. 3a and 3b show the rate of various covariates before and after stratification. Various comorbidities of the disorders are listed in the left column. Before balancing, the odds of psychiatric illness were significantly different among patients with and without neurological disorders (Fig. 3a). After covariate balancing, none of the covariates had a statistically significant difference, showing that cases with neurological disorders had the same rate as controls
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
4
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
Fig. 2. A Network Model for Effect of Diagnoses on Response to Citalopram.
without neurological disorders. Similarly, after covariate balancing for exposure to PTSD, the rate did not differ for any of the covariates for cases exposed to PTSD and controls not exposed to PTSD (alpha < 0.05). These data confirm the accuracy of the covariate balancing in removing the effects of comorbidities as well as the wisdom of stratifying only variables in the Markov Blanket of exposure and the response to antidepressant. After balancing the covariates, the results in Table 2 were observed. The top part of Table 2 shows the results for exposure to neurological disorders. Data reported in this table are the unconfounded odds ratio of response to citalopram. The MantelHaenszel chi-square test of significance shows that response to citalopram is statistically significant. Cases with neurological disorders were compared to controls without neurological disorders. The odds of response was 0.74 (Chi-square = 11.80, significant at alpha < 0.05). We stratified psychiatric illness in the Markov Blanket of PTSD and Neurological disorders in the Markov Blanket of remission to understand the impact of exposure to PTSD. The bottom of Table 2 displays the final results. The un-confounded odds of response for patients exposed to PTSD was 0.53 (Chi-square = 8.31, significant at alpha < 0.05).
Sensitivity analysis The Stratified Covariate Balancing technique matches cases to controls. As the number of comorbidities increases, fewer cases are matched. To verify the sensitivity of the findings to the matching procedure, we also eliminated comorbidities one at a time and repeated the analysis. This provides strata that are partial matches to the comorbidities. Results are presented in Table 3.
Limitations This study had several limitations. First, it is well known that other comorbidities besides the ones balanced in this study (e.g. hypertension or cardiac disease [51]) could affect response to an antidepressant. A more comprehensive approach to controlling variations in comorbidities is needed. Many of these comorbidities are collected during the initial intakes and evaluations conducted on new patients at healthcare clinics and entered into electronic health records. Future studies controlling for a greater number of comorbidities on depression would potentially be able to offer clinicians valuable guideline recommendations. In this paper we relied on stratified covariate balancing to remove the effects of comorbidities in an observational study, in which all patients had received citalopram. This approach does not remove the effects of covariates that were not measured. A randomized clinical study would do so; but randomized studies typically exclude patients with multiple comorbidities, which impair the generalizability of these studies. The advantage of the current study is that it included and statistically controlled for multiple concurrent comorbidities of PTSD or neurological disorders. Further studies are necessary to replicate these findings and to insure that the study conclusions are not unique to the STAR*D database. In addition, while the current study ruled out citalopram as beneficial for some patients, it did not address which medication would be of greatest benefit for these patients. Additional studies are needed comparing the effectiveness of various treatment options for patients with PTSD and neurological disorders. Discussion This study demonstrated the use of a novel analytical method, Stratified Covariate Balancing, for personalizing data. This
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
5
Fig. 3. Odds of Comorbidities in Different Sub-Groups.
approach calculates the personalized odds of remission by stratifying all comorbidities in the data that are not in the patient’s history. We stratified only those comorbidities on the Markov Blanket of remission to reduce the number of stratifications. Variables on the Markov Blanket of the outcome statistically remove the effect of all other comorbidities on the outcome. The procedure allows measurement of the un-confounded, personalized impact of a person’s medical history on health outcomes. In this study we reported the odds ratio of response to citalopram in patients with a history of different mental disorders. In each disorder, we removed the impact of gender, age, race, and 18 comorbidities. The approach ensured that the comorbidities occurred at the same rate in the cases exposed to the disorders as in the controls without the disorders. The common odds ratio among the cases and controls was below 1 and statistically signif-
icant for patients with neurological and post-traumatic stress disorders, indicating that these patients were less likely to experience depression remission after the use of citalopram. These data suggest that citalopram should not be a first line agent prescribed for patients with these two disorders. Other investigators have also discussed the use of citalopram in the treatment of PTSD. Some investigators have pointed out that the use of antidepressants to treat patients with PTSD is common [52–54]. There are mixed results on which antidepressants work best for veterans with war-related PTSD [55]. To our knowledge, sertraline and paroxetine are currently the only antidepressants shown to work for patients with PTSD in double-blind randomized clinical studies [56–62]. These studies do not distinguish between solo PTSD versus PTSD with head trauma due to blast injuries. Head trauma often leads to a diagnosis of neurological disorders.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
6
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
Table 2 Response to Citalopram for Patients with PTSD or Neurological Disorders. Strata
Cases with Neurological Disorder
Depression only Depression, PTSD Strata
*
Controls without Neurological Disorder
Number of Patients
Probability of Response
Number of Patients
Probability of Response
Weight
692 48
0.46 0.31
2053 110
0.53 0.37
0.34 0.44
Cases with PTSD
Controls without PTSD
Number of Patients
Probability of Response
Number of Patients
Probability of Response
Weight
Depression only Depression, Neurological Disorder Depression, Psychiatric Illness Depression, Psychiatric Illness, Neurological Disorder
15 7 95 41
0.27 0.43 0.39 0.29
1759 562 294 130
0.54 0.46 0.50 0.44
0.01 0.01 0.32 0.32
Test of Significance
Common Odds Ratio
Disorders
Cases
Remission
X2
Odds Ratio
Lower/Upper
Neurological Disorder Post-Traumatic Stress Disorder
740 158
332 56
11.80* 8.31*
0.74 0.53
0.69/1.22 0.47/1.22
X2 > 3.841 Significant at alpha 0.05.
Table 3 Partial Matches to Cases with Neurological Disorder and PTSD. Covariate Removed
740 cases with Neurological Disorder Odds Ratio (n)
158 cases with PTSD Odds Ratio (n)
None Age Gender Heart Vascular Hematopoietic Eyes Gastrointestinal Renal Genitourinary Musculoskeletal Respiratory Liver Endocrine Alcohol Cannabis Panic General Anxiety Race
0.26 0.16 0.18 0.34 0.39 0.37 0.33 0.31 0.27 0.27 0.27 0.33 0.30 0.26 0.26 0.28 0.29 0.31 0.32
0.0 (18) 0.0 (29) 0.0 (39) 0.10 (24) 0.14 (29) 0.17 (34) 0.21 (46) 0.46 (63) 0.50 (68) 0.44 (82) 0.53 (92) 0.48 (149) 0.49 (151) 0.49 (154) 0.48 (154) 0.48 (155) 0.53 (155) 0.51 (157) 0.55 (158)
(274) (351) (399) (315) (365) (396) (456) (511) (547) (594) (632) (694) (705) (719) (722) (727) (729) (739) (740)
Our findings controlled for neurological comorbidities of PTSD and thus extended the findings in the literature by suggesting cialopram may be less effective in the subgroup of patients.
References [1] Rush AJ, Wisniewski SR, Warden D, Luther JF, Davis LL, Fava M, et al. Selecting among second-step antidepressant medication monotherapies: predictive value of clinical, demographic, or first-step treatment features. Arch. Gen. Psychiatry. 2008;65(8):870–80. [2] Rost K, Nutting P, Smith JL, Elliott CE, Dickinson M. Managing depression as a chronic disease: a randomised trial of ongoing treatment in primary care. BMJ 2002;325(7370):934. [3] Cipriani A, Furukawa TA, Salanti G, Geddes JR, Higgins JP, Churchill R, et al. Comparative efficacy and acceptability of 12 new-generation antidepressants: a multiple-treatments meta-analysis. Lancet 2009;373(9665):746–58. [4] Cipriani A, Brambilla P, Furukawa T, Geddes J, Gregis M, Hotopf M, Malvini L, Barbui C. Fluoxetine versus other types of pharmacotherapy for depression. Cochrane Database Syst Rev. 2005(4). CD004185. [5] Insel TR, Wang PS. The STAR*D trial: revealing the need for better treatments. Psychiatric Services 2009;60(11):1466–7. [6] Pratt L, Brody DJ, Gu Q. Antidepressant use in persons aged 12 and over: United States, 2005–2008. NCHS Data Brief October 2011;76. [7] Valenstein M, Vijan S, Zeber JE, Boehm K, Buttar A. The cost-utility of screening for depression in primary care. Ann Intern Med 2001;134:345–60.
[8] Howland RH. Sequenced treatment alternatives to relieve depression (STAR*D). Part 2: study outcomes. J Psychosoc Nurs Ment Health Serv 2008;46(10):21–4. [9] Stewart WF, Ricci JA, Chee E, Hahn SR, Morganstein D. Cost of lost productive work time among US workers with depression. J Am Med Assoc 2003;289 (23):3135–44. [10] American Psychiatric Association. Practice Guidelines for the Treatment of Patients with Major Depressive Disorder, 2010. Retrieved from http:// psychiatryonline.org/pb/assets/raw/sitewide/practice_guidelines/guidelines/ mdd.pdf. [11] Montgomery SA, Djärv L. The antidepressant efficacy of citalopram. Int Clin Psychopharmacol 1996;11(Suppl 1):29–33. [12] Arean PA. Personalizing behavioral interventions: the case of late-life depression. Neuropsychiatry (London) 2012;2(2):135–45. [13] Jain FA, Hunter AM, Leuchter AF, Brooks 3rd JO. A role for profiles of patientspecific depression characteristics and socioeconomic factors in the prediction of antidepressant treatment outcome. J Clin Psychiatry 2015;76(3):327. [14] Gilmer WS, Gollan JK, Wisniewski SR, Howland RH, Trivedi MH, Miyahara S, Fleck J, Thase ME, Alpert JE, Nierenberg AA, Warden D, Fava M, Rush AJ. Does the duration of index episode affect the treatment outcome of major depressive disorder? A STAR*D report. J Clin Psychiatry 2008;69(8):1246–56. [15] Rush AJ, Trivedi MH, Wisniewski SR, Nierenberg AA, Stewart JW, Warden D, et al. Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR⁄D report. Am J Psychiatry 2006;163 (11):1905–17. [16] Foulds JA, Douglas Sellman J, Adamson SJ, Boden JM, Mulder RT, Joyce PR. Depression outcome in alcohol dependent patients: an evaluation of the role of independent and substance-induced depression and other predictors. J Affect Disord 2015;15(174):503–10. [17] Howland RH, Rush AJ, Wisniewski SR, Trivedi MH, Warden D, Fava M, et al. Concurrent anxiety and substance use disorders among outpatients with major depression: clinical features and effect on treatment outcome. Drug Alcohol Depend 2009;99(1–3):248–60. [18] Mansoor B, Rengasamy M, Hilton R, Porta G, He J, Spirito A, et al. The bidirectional relationship between body mass index and treatment outcome in adolescents with treatment-resistant depression. J Child Adolescent Psychopharmacology 2013;23(7):458–67. [19] Troxel WM, Kupfer DJ, Reynolds 3rd CF, Frank E, Thase ME, Miewald JM, et al. Insomnia and objectively measured sleep disturbances predict treatment outcome in depressed patients treated with psychotherapy or psychotherapypharmacotherapy combinations. J Clin Psychiatry 2012;73(4):478–85. [20] Miller MD, Lenze EJ, Dew MA, Whyte E, Weber E, Begley AE, Reynolds 3rd CF. Effect of cerebrovascular risk factors on depression treatment outcome in later life. Am J Geriatric Psychiatry 2002;10(5):592–8. [21] Jain FA, Cook IA, Leuchter AF, Hunter AM, Davydov DM, Ottaviani C, et al. Heart rate variability and treatment outcome in major depression: a pilot study. Int J Psychophysiol 2014;93(2):204–10. [22] Chang HS, Won E, Lee HY, Ham BJ, Lee MS. Association analysis for corticotropin releasing hormone polymorphisms with the risk of major depressive disorder and the response to antidepressants. Behav Brain Res 2015;6(292):116–24. [23] Duval F, Mokrani MC, Erb A, Gonzalez Lopera F, Alexa C, Proudnikova X, et al. Chronobiological hypothalamic-pituitary-thyroid axis status and antidepressant outcome in major depression. Psychoneuroendocrinology 2015;59:71–80. [24] Graziottin A, Serafini A. Depression and the menopause: why antidepressants are not enough? Menopause Int 2009;15(2):76–81.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx [25] Markowitz JC, Meehan KB, Petkova E, Zhao Y, Van Meter PE, Neria Y, et al. Treatment preferences of psychotherapy patients with chronic PTSD. J Clin Psychiatry 2015. [26] Nilsen TS, Eisemann M, Kvernmo S. Predictors and moderators of outcome in child and adolescent anxiety and depression: a systematic review of psychological treatment studies. Eur Child Adolescent Psychiatry 2013;22 (2):69–87. [27] Mojtabai R, Olfson M. Proportion of antidepressants prescribed without a psychiatric diagnosis is growing. Health Aff 2010;30(8):1434–42. [28] Winner JG, Carhart JM, Altar CA, Goldfarb S, Allen JD, Lavezzari G, et al. Combinatorial pharmacogenomic guidance for psychiatric medications reduces overall pharmacy costs in a 1 year prospective evaluation. Curr Med Res Opin 2015;31(9):1633–43. [29] Herbild L, Andersen SE, Werge T, Rasmussen HB, Jürgens G. Does pharmacogenetic testing for CYP450 2D6 and 2C19 among patients with diagnoses within the schizophrenic spectrum reduce treatment costs? Basic Clin Pharmacol Toxicol 2013;113(4):266–72. [30] Alemi F, Zargoush M, Erdman H, Vang J, Epstein S, Ayman F. Genetic markers anticipate response to citalopram in a majority of patients. Psychiatr Genet 2011;21(6):287–93. [31] Bühler J, Seemüller F, Läge D. The predictive power of subgroups: an empirical approach to identify depressive symptom patterns that predict response to treatment. J Affect Disord 2014;163:81–7. [32] Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70:41–55. [33] Rosenbaum PR, Rubin DB. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am Statistn 1985;39:33–8. [34] Rosenbaum PR. Optimal matching for observational studies. J Am Statist Ass 1989;84:1024–32. [35] Abadie A, Imbens GW. Large sample properties of matching estimators for average treatment effects. Econometrica 2006;74:235–67. [36] Rosenbaum PR, Rubin DB. Reducing bias in observational studies using subclassification on the propensity score. J Am Statist Ass 1984;79:516–24. [37] Rosenbaum PR. A characterization of optimal designs for observational studies. J R Statist Soc B 1991;53:597–610. [38] Hansen BB. Full matching in an observational study of coaching for the SAT. J Am Statist Ass 2004;99:609–18. [39] Rosenbaum PR. Model-based direct adjustment. J Am Statist Ass 1987;82:387–94. [40] Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000;11:550–60. [41] Hirano K, Imbens G, Ridder G. Efficient estimation of average treatment effects using the estimated propensity score. Econometrica 2003;71:1307–38. [42] Heckman JJ, Ichimura H, Todd P. Matching as an econometric evaluation estimator. Rev. Econ. Stud. 1998;65:261–94. [43] Imai K, Ratkovic M. Covariate balancing propensity score. J R Statist Soc B 2014;76(Part 1):243–63. [44] Robins JM, Rotnitzky A, Zhao LP. Analysis of semiparametric regression models for repeated outcomes in the presence of missing data. J Am Statist Ass 1995;90:106–21.
7
[45] Ho DE, Imai K, King G, Stuart EA. Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Polit Anal 2007;15:199–236. [46] Abadie A, Imbens GW. Bias-corrected matching estimators for average treatment effects. J Bus Econ Statist 2011;29:1–11. [47] Alemi F, Elrafey AH, Avramovic I. Covariate Balancing through Naturally Occurring Strata. In review and available through first author [email protected]. [48] Pearl J, Glymour M, Jewell NP. Causal inference in statistics: a premier. John Wiley and Sons; 2016. [49] Aliferis CF, Tsamardinos I, Statnikov A. HITON: a novel Markov Blanket algorithm for optimal variable selection. AMIA Annu Symp Proc 2003;21–5. [50] Alemi F, Zargoush M, Vang J. Using observed sequence to orient causal networks. Health Care Manag Sci 2016. [51] Reynolds 3rd CF, Dew MA, Pollock BG, Mulsant BH, Frank E, Miller MD, et al. Maintenance treatment of major depression in old age. N Engl J Med 2006;354 (11):1130–8. [52] Tsunashima K, Kato N, Sakikawa N, Ishibashi D. Investigation of drug therapy for posttraumatic stress disorder in Japan. Ther Res 2005;26:1901–9. [53] Shigemura J, Maeda M, Misari O, Kato H, Kameoka T, et al. Survey on pharmacological treatment status of posttraumatic stress disorder after largescale disasters: a multicenter retrospective survey. Jpn J Trauma Stress 2013;11:51–62. [54] Oe M, Maeda M, Kim Y. Prescribing patterns for Japanese PTSD patients using paroxetine. Jpn J Trauma Stress 2007;5:167–74. [55] Ursano RJ, Bell C, Eth S, Friedman M, Norwood A, Pfefferbaum B, et al. Work Group on ASD and PTSD.; Steering Committee on Practice Guidelines. Practice guideline for the treatment of patients with acute stress disorder and posttraumatic stress disorder. Am J Psychiatry 2004;161(11 Suppl):3–31. [56] Pull CN, Pull CB. Current status of treatment for posttraumatic stress disorder: focus on treatments combining pharmacotherapy and cognitive behavioral therapy. Int J Cogn Ther 2014;7(2):149–61. [57] Ipser JC, Stein DJ. Evidence-based pharmacotherapy of post-traumatic stress disorder (PTSD). Int J Neuropsychopharmacol 2012;15:825–40. [58] Davidson JR, Rothbaum BO, van der Kolk BA, Sikes CR, Farfel GM. Multicenter, double-blind comparison of sertraline and placebo in the treatment of posttraumatic stress disorder. Arch Gen Psychiatry 2001;58:485–92. [59] Brady K, Pearlstein T, Asnis GM, Baker D, Rothbaum B, et al. Efficacy and safety of sertraline treatment of posttraumatic stress disorder: a randomized controlled trial. JAMA 2000;283:1837–44. [60] Stein DJ, Davidson J, Seedat S, Beebe K. Paroxetine in the treatment of posttraumatic stress disorder: pooled analysis of placebo-controlled studies. Expert Opin Pharmacother 2003;4:1829–38. [61] Tucker P, Zaninelli R, Yehuda R, Ruggiero L, Dillingham K, et al. Paroxetine in the treatment of chronic posttraumatic stress disorder: results of a placebocontrolled, flexible-dosage trial. J Clin Psychiatry 2001;62:860–8. http://dx. doi.org/10.4088/JCP.v62n1105. [62] Marshall RD, Beebe KK, Oldham M, Zaninelli R. Efficacy and safety of paroxetine treatment for chronic PTSD: a fixed-dose, placebo-controlled study. Am J Psychiatry 2001;158:1982–8. http://dx.doi.org/10.1176/appi. ajp.158.12.1982.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
Contents lists available at ScienceDirect
Personalized medicine in psychiatry journal homepage: www.elsevier.com/locate/pmip
Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder Farrokh Alemi a,⇑, Aryan Mazloum-Yazdi a, Lora Peppard b a b
George Mason University, Department of Health Administration and Policy, United States George Mason University, School of Nursing, United States
a r t i c l e
i n f o
Article history: Received 22 May 2017 Accepted 15 July 2017 Available online xxxx
a b s t r a c t Importance: Citalopram is a common antidepressant widely used and STAR*D study has indicated that most, more than 60%, of patients who receive it, do not benefit from their first antidepressant. Objective: To identify whether citalopram is effective for patients with particular medical history. Design: We used stratification to balance case/control study design. Setting: We used the National Institute of Mental Health’s STAR*D data, collected from 41 specialty and primary medical care settings spanning across a seven year time period. Participants: The data included .mental and physical diagnoses of 4041 patients with major depression. Intervention(s) (for clinical trials) or exposure(s) (for observational studies): Patients comorbidities were balanced using Stratified Covariate Balancing method. This algorithm uses Markov Blanket of both treatment and outcome to identify which comorbidities can be ignored. It uses stratification to remove confounding introduced by the covariates in the Markov Blanket of treatment and outcome. Main outcome(s) and measure(s): A patient’s remission was judged by weather there was a 50% reduction in the Hamilton Rating Scale for Depression. Results: Patients exposed to neurological disorders were less likely to experience remission than controls without neurological disorder (odds = 0.74, chi-square=, alpha < 0.05). For exposure to PTSD, we stratified psychiatric illness in the Markov Blanket of PTSD and Neurological disorders in the Markov Blanket of Remission. The un-confounded odds of response for patients exposed to PTSD was 0.53 (Chi-square = 8.31, alpha < 0.05.) Conclusions and relevance: Patients exposed to PTSD or neurological disorders were less likely to experience remission from depression. These data suggest that citalopram should not be a first line agent prescribed for patients with these disorders. Ó 2017 Elsevier Inc. All rights reserved.
Background More than 60% of depressed patients do not benefit from their first antidepressant [1–4], prompting some experts to argue that ‘‘the bar for antidepressants has been set far too low” [5]. To understand the magnitude of the problem, consider that of the 11% of Americans currently taking antidepressants [6,7], 25 million initially took an antidepressant that was not therapeutic. Months later, after repeated trials on antidepressants, more than 50% of the initial non-respondentseventually benefited from a different antidepressant [8]; indicating that these patients are not treatment resistant and more careful selection of an antidepressant could
⇑ Corresponding author at: 4400 University Drive, Fairfax, VA 22030, United States. E-mail addresses: [email protected] (F. Alemi), [email protected] (A. Mazloum-Yazdi).
provide an earlier response. During the months of mistrial of antidepressants, patients continued to suffer from debilitating symptoms, many remained at risk of suicide and millions of dollars were spend on unnecessary medication [9]. Unfortunately, clinicians have limited guidance on which antidepressant might be the most effective for patients with different medical and psychiatric histories. Guidelines and previous studies have focused on the effectiveness of citalopram and not on characteristics of those who may or may not respond to this medication [10,11]. This paper provides clinicians with additional guidance on who would most likely respond to citalopram, a commonly prescribed antidepressant and one of the first selective serotonin re-uptake inhibitors. The task of choosing an initial antidepressant is complex. Guidelines currently recommend basing the decision on the anticipated side effects, the safety or tolerability of these side effects for the individual patient, pharmacological properties of the medication, medication response in prior episodes, cost, and patient pref-
http://dx.doi.org/10.1016/j.pmip.2017.07.003 2468-1717/Ó 2017 Elsevier Inc. All rights reserved.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
2
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
erence [10]. While a great deal of information is available in numerous randomized clinical studies [11,12], these studies have excluded patients with major comorbidities, making it difficult to anticipate response to treatment for patients with comorbidities. A patient’s race, gender, employment, education, and income have been shown to affect response to treatment [13]. Length of depression episode [14], cognition [22], and the presence of the following have also been shown to affect response to antidepressants: concurrent anxiety [15], substance use disorders [16,17], obesity [18], insomnia [19], cerebrovascular diseases [20,21], hormone imbalances [22–24], and post-traumatic stress disorder [25]. Many of these predictors do not have a consistent impact on response to treatment, making it more difficult for clinicians to decide among various treatment alternatives [26]. In addition, there are more than 24 types of antidepressants currently on the market [27], and many of them have several different brand names and can be prescribed in different dosages, further complicating the choice of the first antidepressant. Attempts to improve the precision of antidepressant selection have had mixed results. Some studies have tried to use genetic markers to predict response with mixed success. One study found these markers work for patients with psychiatric disorders [28,29]. Another study found that genetic markers could accurately (78% correct) predict response in 2/3rd of cases but not all cases [30]. Others have tried to improve the precision of antidepressant selection by dividing depression into sub-types. One study identified 5 different symptom profiles predictive of response to antidepressants [31]. Unfortunately, these profiles did not map to patients’ diagnoses (with the exception of melancholic depression), raising the possibility that these profiles were statistically significant but not clinically useful. In this paper, we examine if diagnoses from the Diagnostic and Statistical Manual of Mental Disorders can be used to predict response to antidepressants. Such an approach is advantageous because clinicians can readily implement the study findings into practice. Methods Study sample We used the STAR*D database, the largest available data set capturing response to antidepressants. These data include genetic markers and phenotypes of 4,041 patients with major depression. Demographic data (gender, age, race), data from 13 disorders (cardiovascular, vascular, hematopoietic, eyes, gastrointestinal, renal, genitourinary, musculoskeletal, neurological, psychiatric, respiratory, liver, and endocrine diagnostic disorders), and verified depression comorbidities (Alcohol, Cannabis, Panic, Posttraumatic Stress, and Generalized Anxiety disorders) were examined in this study. Data were collected from 41 clinical sites around the country, which included both specialty care settings and primary medical care settings, over a seven year time period. The response to citalopram was measured using the Hamilton Rating Scale for Depression (HRSD). Patient remission was recognized as a 50% decrease in HRSD score from the start of the trial. Method of analysis In order to accurately predict the response to citalopram for patients having a particular diagnosis, the influence of patient comorbidities should be accounted for. Until recently statistical approaches to control for comorbidities were not available. In 1983, Rosenbaum and Rubin proposed the use of propensity scoring to balance rates of occurrence of covariates/comorbidities [32]. Since then, the approach has been revised and used widely [33–
46]. We used Stratified Covariate Balancing, where the propensity weights are derived analytically without statistical modeling and through stratification [47]. All patients had received citalopram and were assigned to cases and controls based on exposure to specific disorders. Thus, in evaluating the impact of exposure to post-traumatic stress disorder (PTSD) on remission, we considered all patients with PTSD as cases, all patients without PTSD as controls, and all other comorbidities as covariates that should be balanced through stratification. One limitation of using stratification to control for effect of comorbidities is that as the number of covariates increases, only a few cases fall within each stratum. A number of methods are available to restrict the stratification to a smaller number of covariates. One approach that has been shown to be theoretically sound is to restrict the analysis to the Markov Blanket of outcome [48,49]. We fit a Bayesian probability network to the data to identify the covariates that are in the Markov Blanket of remission and them identified a temporal sequence between any two diagnoses using age at which the patients reported the diagnosis to accurately fit the network. The use of temporal sequence among a pair of variables to improve accuracy of network modeling has been reported elsewhere [50]. Diagnoses that occur later were prohibited from affecting earlier events. The BNLearn software package in R version (3.31) was used to fit the network model. Once the Markov Blanket for both (a) the exposure variable and (b) remission variables were identified, these covariates were stratified and the un-confounded impact of exposure on remission reported.
Results Table 1 shows the demographics and comorbidities of patients in our sample. To show how patients exposed to specific disorders have different comorbidities and demographics, Table 1 reports the data for exposure to PTSD and neurological disorders. Patients exposed to PTSD differed from patients not exposed to PTSD in nineteen covariates. Patients exposed to neurological disorders also differed from patients not exposed to neurological disorders in twenty covariates. These data show significant variations in demographics and comorbidities. The impact of exposure to either PTSD or to neurological disorders on response to citalopram cannot be estimated without statistically controlling for the listed comorbidities. We calculated the average age at which each diagnoses occurs to understand what variables are in the Markov Blanket of exposure and outcome. Age, race, and gender are events that occur at birth. The outcome variable occurs at end of the data collection effort and therefore no disorders can occur after it. Fig. 1 provides the age at which various diagnoses are given. For example, at alpha levels less than 0.05, cannabis abuse (Average Age = 411.30 months, Standard Deviation = 130.73 months) occurs prior to heart disease (Average = 589.76 months, standard deviation = 157.49 months). Based on these data, we organized the diagnoses into 8 categories as follows: Time 1: Age, Gender, Race Time 2: Panic Disorder, Cannabis Abuse Time 3: Neurological Disorder, General Anxiety Disorder, Psychiatric Illness, Alcohol Abuse Time 4: Gastrointestinal Disorder, Respiratory Disease, Hematopoietic Disorder, Eyes/Ears/Nose/Throat/Larynx Disorder Time 5: Musculoskeletal/Integument Disorder, Genitourinary Disorder, Renal Disease Time 6: Heart Disease, Endocrine Disorder, Liver Disease, PostTraumatic Stress Disorder Time 7: Vascular Disease Time 8: Response to Citalopram
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
3
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx Table 1 Rate of Various Events in STAR*D Data before Covariate Balancing. Neurological Disorder Cases Age 41+ Age 18–41 Gender Female Gender Male Race White Race Not-White Comorbidities
No comorbidities Heart Vascular Hematopoietic Eyes Gastrointestinal Renal Genitourinary Musculoskeletal Psychiatric Illness Respiratory Liver Endocrine Alcohol Abuse Cannabis Panic Generalized Anxiety Total
395 345 491* 249* 557 183 24** 145** 218** 81* 442** 397** 65** 268** 417** 171** 319** 102** 245** 43 12 38 51 740
Controls without Neurological Disorder 1102 1061 1299* 864* 1726 437 300** 595** 482** 160* 841** 694** 89** 417** 850** 389** 595** 149** 452** 110 31 86 114 2163
PTSD Cases **
115 43** 57** 101** 100** 58** 1** 40** 62** 15 98** 70 12 59** 110** 136** 70** 16 46 13 3 18** 13 158
Controls without PTSD 1382** 1363** 1733** 1012** 2183** 562** 324** 386** 638** 226 1185** 1021 142 626** 1157** 424** 844** 235 651 140 40 106** 152 2745
* Statistically significant at 0.05, ** Statistically significant at 0.01.
Fig. 1. Age of Occurrence of Various Diseases. Cell entries indicate > if row variables occurs after and < if it occurs before the column variable. Blank entries indicate pairs of diseases with no significant difference in age of occurrence.
In fitting a network model to the data, we prohibited the algorithm from allowing a later event to influence an earlier event. The resulting network structure is provided in Fig. 2. Certain variables (specific phobia, obsessive compulsive disorder, other disorders, and ethnicity) were not connected to any member of the network and were dropped from the analysis. The network shows the Markov Blanket of response to citalopram. The Markov Blanket is the set of variables that directionally separate the network from response to citalopram. This includes two variables: Neurological disorder and PTSD. Furthermore, the Markov Blanket of exposure to PTSD includes Psychiatric Illness. To find the effect of PTSD on
response after taking citalopram, we need to stratify the Markov Blanket of PTSD as well as the Markov Blanket of response; in short we need to only stratify two (Neurological Disorder and PTSD) of the twenty-one possible covariates. Fig. 3a and 3b show the rate of various covariates before and after stratification. Various comorbidities of the disorders are listed in the left column. Before balancing, the odds of psychiatric illness were significantly different among patients with and without neurological disorders (Fig. 3a). After covariate balancing, none of the covariates had a statistically significant difference, showing that cases with neurological disorders had the same rate as controls
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
4
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
Fig. 2. A Network Model for Effect of Diagnoses on Response to Citalopram.
without neurological disorders. Similarly, after covariate balancing for exposure to PTSD, the rate did not differ for any of the covariates for cases exposed to PTSD and controls not exposed to PTSD (alpha < 0.05). These data confirm the accuracy of the covariate balancing in removing the effects of comorbidities as well as the wisdom of stratifying only variables in the Markov Blanket of exposure and the response to antidepressant. After balancing the covariates, the results in Table 2 were observed. The top part of Table 2 shows the results for exposure to neurological disorders. Data reported in this table are the unconfounded odds ratio of response to citalopram. The MantelHaenszel chi-square test of significance shows that response to citalopram is statistically significant. Cases with neurological disorders were compared to controls without neurological disorders. The odds of response was 0.74 (Chi-square = 11.80, significant at alpha < 0.05). We stratified psychiatric illness in the Markov Blanket of PTSD and Neurological disorders in the Markov Blanket of remission to understand the impact of exposure to PTSD. The bottom of Table 2 displays the final results. The un-confounded odds of response for patients exposed to PTSD was 0.53 (Chi-square = 8.31, significant at alpha < 0.05).
Sensitivity analysis The Stratified Covariate Balancing technique matches cases to controls. As the number of comorbidities increases, fewer cases are matched. To verify the sensitivity of the findings to the matching procedure, we also eliminated comorbidities one at a time and repeated the analysis. This provides strata that are partial matches to the comorbidities. Results are presented in Table 3.
Limitations This study had several limitations. First, it is well known that other comorbidities besides the ones balanced in this study (e.g. hypertension or cardiac disease [51]) could affect response to an antidepressant. A more comprehensive approach to controlling variations in comorbidities is needed. Many of these comorbidities are collected during the initial intakes and evaluations conducted on new patients at healthcare clinics and entered into electronic health records. Future studies controlling for a greater number of comorbidities on depression would potentially be able to offer clinicians valuable guideline recommendations. In this paper we relied on stratified covariate balancing to remove the effects of comorbidities in an observational study, in which all patients had received citalopram. This approach does not remove the effects of covariates that were not measured. A randomized clinical study would do so; but randomized studies typically exclude patients with multiple comorbidities, which impair the generalizability of these studies. The advantage of the current study is that it included and statistically controlled for multiple concurrent comorbidities of PTSD or neurological disorders. Further studies are necessary to replicate these findings and to insure that the study conclusions are not unique to the STAR*D database. In addition, while the current study ruled out citalopram as beneficial for some patients, it did not address which medication would be of greatest benefit for these patients. Additional studies are needed comparing the effectiveness of various treatment options for patients with PTSD and neurological disorders. Discussion This study demonstrated the use of a novel analytical method, Stratified Covariate Balancing, for personalizing data. This
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
5
Fig. 3. Odds of Comorbidities in Different Sub-Groups.
approach calculates the personalized odds of remission by stratifying all comorbidities in the data that are not in the patient’s history. We stratified only those comorbidities on the Markov Blanket of remission to reduce the number of stratifications. Variables on the Markov Blanket of the outcome statistically remove the effect of all other comorbidities on the outcome. The procedure allows measurement of the un-confounded, personalized impact of a person’s medical history on health outcomes. In this study we reported the odds ratio of response to citalopram in patients with a history of different mental disorders. In each disorder, we removed the impact of gender, age, race, and 18 comorbidities. The approach ensured that the comorbidities occurred at the same rate in the cases exposed to the disorders as in the controls without the disorders. The common odds ratio among the cases and controls was below 1 and statistically signif-
icant for patients with neurological and post-traumatic stress disorders, indicating that these patients were less likely to experience depression remission after the use of citalopram. These data suggest that citalopram should not be a first line agent prescribed for patients with these two disorders. Other investigators have also discussed the use of citalopram in the treatment of PTSD. Some investigators have pointed out that the use of antidepressants to treat patients with PTSD is common [52–54]. There are mixed results on which antidepressants work best for veterans with war-related PTSD [55]. To our knowledge, sertraline and paroxetine are currently the only antidepressants shown to work for patients with PTSD in double-blind randomized clinical studies [56–62]. These studies do not distinguish between solo PTSD versus PTSD with head trauma due to blast injuries. Head trauma often leads to a diagnosis of neurological disorders.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
6
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx
Table 2 Response to Citalopram for Patients with PTSD or Neurological Disorders. Strata
Cases with Neurological Disorder
Depression only Depression, PTSD Strata
*
Controls without Neurological Disorder
Number of Patients
Probability of Response
Number of Patients
Probability of Response
Weight
692 48
0.46 0.31
2053 110
0.53 0.37
0.34 0.44
Cases with PTSD
Controls without PTSD
Number of Patients
Probability of Response
Number of Patients
Probability of Response
Weight
Depression only Depression, Neurological Disorder Depression, Psychiatric Illness Depression, Psychiatric Illness, Neurological Disorder
15 7 95 41
0.27 0.43 0.39 0.29
1759 562 294 130
0.54 0.46 0.50 0.44
0.01 0.01 0.32 0.32
Test of Significance
Common Odds Ratio
Disorders
Cases
Remission
X2
Odds Ratio
Lower/Upper
Neurological Disorder Post-Traumatic Stress Disorder
740 158
332 56
11.80* 8.31*
0.74 0.53
0.69/1.22 0.47/1.22
X2 > 3.841 Significant at alpha 0.05.
Table 3 Partial Matches to Cases with Neurological Disorder and PTSD. Covariate Removed
740 cases with Neurological Disorder Odds Ratio (n)
158 cases with PTSD Odds Ratio (n)
None Age Gender Heart Vascular Hematopoietic Eyes Gastrointestinal Renal Genitourinary Musculoskeletal Respiratory Liver Endocrine Alcohol Cannabis Panic General Anxiety Race
0.26 0.16 0.18 0.34 0.39 0.37 0.33 0.31 0.27 0.27 0.27 0.33 0.30 0.26 0.26 0.28 0.29 0.31 0.32
0.0 (18) 0.0 (29) 0.0 (39) 0.10 (24) 0.14 (29) 0.17 (34) 0.21 (46) 0.46 (63) 0.50 (68) 0.44 (82) 0.53 (92) 0.48 (149) 0.49 (151) 0.49 (154) 0.48 (154) 0.48 (155) 0.53 (155) 0.51 (157) 0.55 (158)
(274) (351) (399) (315) (365) (396) (456) (511) (547) (594) (632) (694) (705) (719) (722) (727) (729) (739) (740)
Our findings controlled for neurological comorbidities of PTSD and thus extended the findings in the literature by suggesting cialopram may be less effective in the subgroup of patients.
References [1] Rush AJ, Wisniewski SR, Warden D, Luther JF, Davis LL, Fava M, et al. Selecting among second-step antidepressant medication monotherapies: predictive value of clinical, demographic, or first-step treatment features. Arch. Gen. Psychiatry. 2008;65(8):870–80. [2] Rost K, Nutting P, Smith JL, Elliott CE, Dickinson M. Managing depression as a chronic disease: a randomised trial of ongoing treatment in primary care. BMJ 2002;325(7370):934. [3] Cipriani A, Furukawa TA, Salanti G, Geddes JR, Higgins JP, Churchill R, et al. Comparative efficacy and acceptability of 12 new-generation antidepressants: a multiple-treatments meta-analysis. Lancet 2009;373(9665):746–58. [4] Cipriani A, Brambilla P, Furukawa T, Geddes J, Gregis M, Hotopf M, Malvini L, Barbui C. Fluoxetine versus other types of pharmacotherapy for depression. Cochrane Database Syst Rev. 2005(4). CD004185. [5] Insel TR, Wang PS. The STAR*D trial: revealing the need for better treatments. Psychiatric Services 2009;60(11):1466–7. [6] Pratt L, Brody DJ, Gu Q. Antidepressant use in persons aged 12 and over: United States, 2005–2008. NCHS Data Brief October 2011;76. [7] Valenstein M, Vijan S, Zeber JE, Boehm K, Buttar A. The cost-utility of screening for depression in primary care. Ann Intern Med 2001;134:345–60.
[8] Howland RH. Sequenced treatment alternatives to relieve depression (STAR*D). Part 2: study outcomes. J Psychosoc Nurs Ment Health Serv 2008;46(10):21–4. [9] Stewart WF, Ricci JA, Chee E, Hahn SR, Morganstein D. Cost of lost productive work time among US workers with depression. J Am Med Assoc 2003;289 (23):3135–44. [10] American Psychiatric Association. Practice Guidelines for the Treatment of Patients with Major Depressive Disorder, 2010. Retrieved from http:// psychiatryonline.org/pb/assets/raw/sitewide/practice_guidelines/guidelines/ mdd.pdf. [11] Montgomery SA, Djärv L. The antidepressant efficacy of citalopram. Int Clin Psychopharmacol 1996;11(Suppl 1):29–33. [12] Arean PA. Personalizing behavioral interventions: the case of late-life depression. Neuropsychiatry (London) 2012;2(2):135–45. [13] Jain FA, Hunter AM, Leuchter AF, Brooks 3rd JO. A role for profiles of patientspecific depression characteristics and socioeconomic factors in the prediction of antidepressant treatment outcome. J Clin Psychiatry 2015;76(3):327. [14] Gilmer WS, Gollan JK, Wisniewski SR, Howland RH, Trivedi MH, Miyahara S, Fleck J, Thase ME, Alpert JE, Nierenberg AA, Warden D, Fava M, Rush AJ. Does the duration of index episode affect the treatment outcome of major depressive disorder? A STAR*D report. J Clin Psychiatry 2008;69(8):1246–56. [15] Rush AJ, Trivedi MH, Wisniewski SR, Nierenberg AA, Stewart JW, Warden D, et al. Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR⁄D report. Am J Psychiatry 2006;163 (11):1905–17. [16] Foulds JA, Douglas Sellman J, Adamson SJ, Boden JM, Mulder RT, Joyce PR. Depression outcome in alcohol dependent patients: an evaluation of the role of independent and substance-induced depression and other predictors. J Affect Disord 2015;15(174):503–10. [17] Howland RH, Rush AJ, Wisniewski SR, Trivedi MH, Warden D, Fava M, et al. Concurrent anxiety and substance use disorders among outpatients with major depression: clinical features and effect on treatment outcome. Drug Alcohol Depend 2009;99(1–3):248–60. [18] Mansoor B, Rengasamy M, Hilton R, Porta G, He J, Spirito A, et al. The bidirectional relationship between body mass index and treatment outcome in adolescents with treatment-resistant depression. J Child Adolescent Psychopharmacology 2013;23(7):458–67. [19] Troxel WM, Kupfer DJ, Reynolds 3rd CF, Frank E, Thase ME, Miewald JM, et al. Insomnia and objectively measured sleep disturbances predict treatment outcome in depressed patients treated with psychotherapy or psychotherapypharmacotherapy combinations. J Clin Psychiatry 2012;73(4):478–85. [20] Miller MD, Lenze EJ, Dew MA, Whyte E, Weber E, Begley AE, Reynolds 3rd CF. Effect of cerebrovascular risk factors on depression treatment outcome in later life. Am J Geriatric Psychiatry 2002;10(5):592–8. [21] Jain FA, Cook IA, Leuchter AF, Hunter AM, Davydov DM, Ottaviani C, et al. Heart rate variability and treatment outcome in major depression: a pilot study. Int J Psychophysiol 2014;93(2):204–10. [22] Chang HS, Won E, Lee HY, Ham BJ, Lee MS. Association analysis for corticotropin releasing hormone polymorphisms with the risk of major depressive disorder and the response to antidepressants. Behav Brain Res 2015;6(292):116–24. [23] Duval F, Mokrani MC, Erb A, Gonzalez Lopera F, Alexa C, Proudnikova X, et al. Chronobiological hypothalamic-pituitary-thyroid axis status and antidepressant outcome in major depression. Psychoneuroendocrinology 2015;59:71–80. [24] Graziottin A, Serafini A. Depression and the menopause: why antidepressants are not enough? Menopause Int 2009;15(2):76–81.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003
F. Alemi et al. / Personalized medicine in psychiatry xxx (2017) xxx–xxx [25] Markowitz JC, Meehan KB, Petkova E, Zhao Y, Van Meter PE, Neria Y, et al. Treatment preferences of psychotherapy patients with chronic PTSD. J Clin Psychiatry 2015. [26] Nilsen TS, Eisemann M, Kvernmo S. Predictors and moderators of outcome in child and adolescent anxiety and depression: a systematic review of psychological treatment studies. Eur Child Adolescent Psychiatry 2013;22 (2):69–87. [27] Mojtabai R, Olfson M. Proportion of antidepressants prescribed without a psychiatric diagnosis is growing. Health Aff 2010;30(8):1434–42. [28] Winner JG, Carhart JM, Altar CA, Goldfarb S, Allen JD, Lavezzari G, et al. Combinatorial pharmacogenomic guidance for psychiatric medications reduces overall pharmacy costs in a 1 year prospective evaluation. Curr Med Res Opin 2015;31(9):1633–43. [29] Herbild L, Andersen SE, Werge T, Rasmussen HB, Jürgens G. Does pharmacogenetic testing for CYP450 2D6 and 2C19 among patients with diagnoses within the schizophrenic spectrum reduce treatment costs? Basic Clin Pharmacol Toxicol 2013;113(4):266–72. [30] Alemi F, Zargoush M, Erdman H, Vang J, Epstein S, Ayman F. Genetic markers anticipate response to citalopram in a majority of patients. Psychiatr Genet 2011;21(6):287–93. [31] Bühler J, Seemüller F, Läge D. The predictive power of subgroups: an empirical approach to identify depressive symptom patterns that predict response to treatment. J Affect Disord 2014;163:81–7. [32] Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70:41–55. [33] Rosenbaum PR, Rubin DB. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am Statistn 1985;39:33–8. [34] Rosenbaum PR. Optimal matching for observational studies. J Am Statist Ass 1989;84:1024–32. [35] Abadie A, Imbens GW. Large sample properties of matching estimators for average treatment effects. Econometrica 2006;74:235–67. [36] Rosenbaum PR, Rubin DB. Reducing bias in observational studies using subclassification on the propensity score. J Am Statist Ass 1984;79:516–24. [37] Rosenbaum PR. A characterization of optimal designs for observational studies. J R Statist Soc B 1991;53:597–610. [38] Hansen BB. Full matching in an observational study of coaching for the SAT. J Am Statist Ass 2004;99:609–18. [39] Rosenbaum PR. Model-based direct adjustment. J Am Statist Ass 1987;82:387–94. [40] Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000;11:550–60. [41] Hirano K, Imbens G, Ridder G. Efficient estimation of average treatment effects using the estimated propensity score. Econometrica 2003;71:1307–38. [42] Heckman JJ, Ichimura H, Todd P. Matching as an econometric evaluation estimator. Rev. Econ. Stud. 1998;65:261–94. [43] Imai K, Ratkovic M. Covariate balancing propensity score. J R Statist Soc B 2014;76(Part 1):243–63. [44] Robins JM, Rotnitzky A, Zhao LP. Analysis of semiparametric regression models for repeated outcomes in the presence of missing data. J Am Statist Ass 1995;90:106–21.
7
[45] Ho DE, Imai K, King G, Stuart EA. Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Polit Anal 2007;15:199–236. [46] Abadie A, Imbens GW. Bias-corrected matching estimators for average treatment effects. J Bus Econ Statist 2011;29:1–11. [47] Alemi F, Elrafey AH, Avramovic I. Covariate Balancing through Naturally Occurring Strata. In review and available through first author [email protected]. [48] Pearl J, Glymour M, Jewell NP. Causal inference in statistics: a premier. John Wiley and Sons; 2016. [49] Aliferis CF, Tsamardinos I, Statnikov A. HITON: a novel Markov Blanket algorithm for optimal variable selection. AMIA Annu Symp Proc 2003;21–5. [50] Alemi F, Zargoush M, Vang J. Using observed sequence to orient causal networks. Health Care Manag Sci 2016. [51] Reynolds 3rd CF, Dew MA, Pollock BG, Mulsant BH, Frank E, Miller MD, et al. Maintenance treatment of major depression in old age. N Engl J Med 2006;354 (11):1130–8. [52] Tsunashima K, Kato N, Sakikawa N, Ishibashi D. Investigation of drug therapy for posttraumatic stress disorder in Japan. Ther Res 2005;26:1901–9. [53] Shigemura J, Maeda M, Misari O, Kato H, Kameoka T, et al. Survey on pharmacological treatment status of posttraumatic stress disorder after largescale disasters: a multicenter retrospective survey. Jpn J Trauma Stress 2013;11:51–62. [54] Oe M, Maeda M, Kim Y. Prescribing patterns for Japanese PTSD patients using paroxetine. Jpn J Trauma Stress 2007;5:167–74. [55] Ursano RJ, Bell C, Eth S, Friedman M, Norwood A, Pfefferbaum B, et al. Work Group on ASD and PTSD.; Steering Committee on Practice Guidelines. Practice guideline for the treatment of patients with acute stress disorder and posttraumatic stress disorder. Am J Psychiatry 2004;161(11 Suppl):3–31. [56] Pull CN, Pull CB. Current status of treatment for posttraumatic stress disorder: focus on treatments combining pharmacotherapy and cognitive behavioral therapy. Int J Cogn Ther 2014;7(2):149–61. [57] Ipser JC, Stein DJ. Evidence-based pharmacotherapy of post-traumatic stress disorder (PTSD). Int J Neuropsychopharmacol 2012;15:825–40. [58] Davidson JR, Rothbaum BO, van der Kolk BA, Sikes CR, Farfel GM. Multicenter, double-blind comparison of sertraline and placebo in the treatment of posttraumatic stress disorder. Arch Gen Psychiatry 2001;58:485–92. [59] Brady K, Pearlstein T, Asnis GM, Baker D, Rothbaum B, et al. Efficacy and safety of sertraline treatment of posttraumatic stress disorder: a randomized controlled trial. JAMA 2000;283:1837–44. [60] Stein DJ, Davidson J, Seedat S, Beebe K. Paroxetine in the treatment of posttraumatic stress disorder: pooled analysis of placebo-controlled studies. Expert Opin Pharmacother 2003;4:1829–38. [61] Tucker P, Zaninelli R, Yehuda R, Ruggiero L, Dillingham K, et al. Paroxetine in the treatment of chronic posttraumatic stress disorder: results of a placebocontrolled, flexible-dosage trial. J Clin Psychiatry 2001;62:860–8. http://dx. doi.org/10.4088/JCP.v62n1105. [62] Marshall RD, Beebe KK, Oldham M, Zaninelli R. Efficacy and safety of paroxetine treatment for chronic PTSD: a fixed-dose, placebo-controlled study. Am J Psychiatry 2001;158:1982–8. http://dx.doi.org/10.1176/appi. ajp.158.12.1982.
Please cite this article in press as: Alemi F et al. Citalopram is less effective for patients with neurological disorder and/or post-traumatic stress disorder. Personalized medicine in psychiatry (2017), http://dx.doi.org/10.1016/j.pmip.2017.07.003