Stevenbagley.net

The Nortriptyline Therapeutic Window: A systematic review Abstract
dress variation in study designs and other study-level variables.
Background: Some studies have found a curvilin-ear relationship between the nortriptyline plasma In thirty-three clinical trials or obser- concentration and the clinical rating of depression vational studies, some evidence was found for a in those treated. The concentration interval of op- plateau or decline effect at higher plasma con- timum response has been called the “nortriptyline centrations, but the consistency of the effect var- ied with the depression scale. There was marked Objectives: To analyze all original data relevant to variation in response within and outside the win- dow, and no sharp cutoff in response at either the Data sources: English language articles found in lower or upper window limits. A random-effects PubMed using search terms “nortriptyline,” and meta-analysis found a statistically significant risk “plasma concentration,” “serum concentration,” or difference for treatment within the window com- “blood concentration”. The search was extended pared to treatment above the window of 0.17 (0.04– through the use of Google Scholar citation links, 0.30, 95% CI), corresponding to a number needed and reference mining of secondary articles and showed stronger response than the clinical trials; Study selection: Clinical trials or observational this depended on a single, anomalous outlier. The studies of adults that reported patient-level or ag- meta-regression showed a non-statistically signifi- gregate results of depression response by concen- cant trend for a stronger response in observational tration or concentration range were retained.
studies, and small and non-significant effects for Data extraction: Plasma concentration and de- studies with endogenous depression and those us- pression ratings were extracted from text, tables and ing fixed-dose regimens. Many studies were small and did not attempt to disconfirm the window’s existence by including patients with concentrations tive results were reported narratively.
tive patient-level and aggregate data were meta-analyzed to compare treatment inside the plasma concentration window with treatment outside the highly varying quality for the nortriptyline thera- window. A meta-regression was performed to ad- peutic window was found. Its clinical utility overprudent prescribing practice is limited.
∗Biomedical Informatics Training Program and Department of Genetics, Stanford University. Contact information: StanfordUniversity School of Medicine, 1265 Welch Road MC 5479, MSOBX- 211, Stanford, CA 94305-5479, p: 650-725-8422, f: 650-725-7944, [email protected]. c 2013, Steven C. Bagley dose-response, systematic review, meta-analysis Introduction
containing the region of most effective response,was named the “nortriptyline therapeutic window”.
The rise of evidence-based medicine to guide clin- Follow-up studies showed mixed results, some sup- ical practice, along with basic research providing porting a curvilinear relationship and others failing novel insights into neuropsychiatric mechanisms to replicate those findings, but belief in the window of action, and the curation and deployment of continues; unqualified statements about it appear in databases of drug-drug, and drug-food interactions, standard psychopharmacology texts [51, 55].
have all contributed to a rigorous evidence base The purpose of this review is to evaluate all that supports the safe and effective use of newly known published evidence relevant to the existence developed pharmaceuticals. However, older drugs, of the nortritypline therapeutic window using a sys- which may still be mainstays of treatment, are not tematic review, meta-analysis and meta-regression.
always retrospectively subjected to the same scien-tific scrutiny as their newer counterparts.
researchers have revisited the evidence supporting older drugs, such as trazodone [39], but this has notbecome routine practice, leaving critical gaps in our Search strategy
The tricyclic antidepressants, named for their The PubMed database was searched for English common chemical structure, have been effectively language articles in the interval 1966 through Au- used for decades in the treatment of depression, and gust 2011, using search terms “nortriptyline” AND remain valuable second-line pharmacologic agents.
(“plasma concentration” OR “serum concentration” By the early 1970’s, a problem of pharmacokinetic variation of the tricyclics had been identified. Pa- were reviewed, and potentially relevant articles tients treated with identical doses of a tricyclic had were retrieved. Included were articles, book chap- differing plasma concentrations of the drug and ters, or case series reporting original data on adults its metabolites; sampling across patients revealed a under acute treatment for depression that related wide range of drug and metabolite concentrations nortriptyline concentration to patient depression re- of up to forty-fold but the response of each patient sponse with corresponding depression rating scores to a given dose was generally consistent and repeat- in at least one arm of treatment. Excluded were able [1]. These differences were presumed to be due individual case reports, as well as articles that in- to genetic determinants of drug metabolism. Thus, volved treatment with amitriptyline, of which nor- it seemed plausible that patients might exhibit more triptyline is a metabolite, in order to avoid con- consistent responses to a specific drug plasma con- founding effects. Using Google Scholar, articles cit- centration than to a specific drug dose.
ing results in the first search group were examined.
This process of expansion through citation links was leagues [3] who reported an unusual pharmaco- iterated until no new articles were found, or until logic feature of one of the tricyclics, nortriptyline.
articles outside the scope of this review were en- They found an unexpected curvilinear relationship countered. Bibliographies of retrieved articles, re- between its plasma level and a patient’s therapeutic views, and psychopharmacology texts and mono- improvement as measured by changes in scores on a standardized depression rating scale. With increas- All studies meeting entry criteria were coded ing nortriptyline plasma levels, clinical response without regard to completeness or methodologi- first improved, but at higher levels the response cal quality following current recommended practice plateaued and ultimately declined. When response [34]; some study characteristics that might influ- was plotted against plasma level, their data de- ence quality were explicitly modeled in the meta- scribed an inverted-U shape; the middle region, Data extraction and coding
the Cr ¨onholm-Ottosson (CO) scale [4], the Hamil-ton Depression Rating Scale (HAMD) [21], and Articles were coded in a standard format [34] in a two-dimensional table, including information on (MADRS) [40]. To convert individual (patient-level) populations, interventions, comparators, outcomes, data into aggregate remission data, scores were and study designs. The population was adults, in- thresholded; patients were assigned remission sta- cluding geriatric patients; age ranges or means were tus if their scores were less than or equal to 3 for CO scores, 7 for HAMD scores, and 9 for MADRS scores The inclusion diagnosis was coded as endogenous [54, 29]. The consensus range of 50–150 ng/mL depression versus other. This choice was motivated was used as the therapeutic window [35]. When by the observation that the window phenomenon aggregate results were reported, the stated depres- might be more easily detectable in endogenously sion cutoffs (or other criteria) for determining re- depressed patients [18]. Futher distinctions were mission (or recovery) were recorded, however, in not useful, because of the wide variety of diagnos- some cases, the reported criterion was not what is tic systems (including the Feighner criteria [15], Re- now accepted as the depression cutoff for remis- search Diagnostic Criteria [58], DSM-III, DSM-IIIR, sion. Patient-level data were plotted along with vi- DSM-IV), and diagnoses (including major depres- sual demarcations of the appropriate remission cut- sion, both psychotic and non-psychotic, bipolar dis- off (shown in the plots as a horizontal rectangle), order, and poorly characterized states such as “pri- the therapeutic window (vertical rectangle), and a mary depressive illness”) employed.
locally weighted regression smoother (loess curve) The intervention was treatment with nortripty- [12] as a non-parametric summary of the overall line. The treatment was coded for dose (or dose range), and the dose regimen: fixed (patient as-signed to a fixed-dose or dose titration schedulethroughout the trial), variable (dose varied in unspe- Data analysis
cific way), level (patient’s dose adjusted to producea predetermined target plasma level), or clinician- The effect size was defined to be the risk differ- determined (dose adjusted to achieve clinical re- ence for recovery in the treatment group comparedsponse).
to recovery in the control group. The treatment Study design was recorded as clinical trial group were those patients whose plasma concentra- when an explicit experimenter-controlled interven- tions were inside the therapeutic window. The con- tion was deployed in a prospective fashion; others trol group were those patients with concentrations were coded as observational studies. When relevant above the window; an alternative control group in- data were a subset of the fully reported data, such cluding patients either above or below the window as when one arm of a clinical trial was treated with nortriptyline, only the data from that subset were Formally, with the following definitions: in is retained. Therefore, the sample sizes used for the the number of subjects with plasma concentrations meta-analysis in some cases vary from the total size within the therapeutic window, inrec is the num- ber of such patients who have depression ratings Individual patient data were taken from the text, in the recovered range, above is the number of sub- tables, and figures, extracting the data points from jects with concentrations above the window’s upper scanned images where necessary. Following current limit, and aboverec is the number of such patients re- practice, outcomes were taken as final depression covered, then the risk difference (RD) effect size is scores were also recorded and analyzed.
Three depression rating scales were encountered: Because of the diversity of populations and study Some problems with the data were uncovered.
designs considered, DerSimonian-Laird random- Asberg et al. [3] reported that one patient made effects modeling [13] was employed. Publication a suicide attempt in the second week of their study, bias was assessed using a funnel plot and the trim “which made formal rating impossible”. However, and fill method of Duval and Tweedie [14]. Hetero- they chose to keep that patient’s data and arbitrar- geneity was computed using the Cochrane Q and ily assigned an amelioration (rating scale change) I2 statistics [7]. Various continuity corrections were score of zero. As this imputation seems unjustified, applied to avoid zero in the denominators in the that point was removed from the data set used here.
risk difference (and also in the odds ratio when Montgomery et al. [41] provided few details about used); typically 0.5 was added to each value. All the design of their study; details of recruitment and of these choices of meta-analysis parameters were inclusion were not reported beyond noting that the varied, as detailed below in the sensitivity analy- patients were depressed and not taking antidepres- sis. The canonical interpretation of the significance sants. Recovery was assessed “globally” at one hos- of statistical tests with a threshold of 0.05 (5%) was pital site and retrospectively by chart review at the employed. The quality of evidence was assessed ac- other. Because of the uncertainty of its study design, it was assigned to the “observational” category.
Software
Studies with only qualitative results
Burrows and colleagues published two trials in The analysis was conducted using the R language, In the first [9], 80 patients with “primary [52]. The metafor package (version 1.6-0) was used depressive illness” were treated with nortriptyline to perform the meta-analysis [60], with additional with doses 75–250 mg per day (as determined by the clinician) for four weeks. They concluded that theirresults showed: “a lack of relationship between clin-ical response and plasma nortriptyline levels”. In the second [10], 40 patients were treated in a se-quential matched-pair design, where one patient of A PubMed search was performed on September the pair was randomly assigned to a low plasma level (below 49 ng/mL) and the other to a high level nal studies on the nortriptyline therapeutic window (above 140 ng/mL), each for four weeks. Their trial, meeting the inclusion criteria were found, and are “showed no differences between these plasma levels shown in Table 1. Three pairs of duplicate publica- and clinical response in twenty pairs of depressed tions were identified: [29, 28], [56, 31], [63, 64]; data from the duplicate studies were merged, leaving 30 Fensbo [16] described a trial in which 23 pa- studies for this review. The 30 studies fall into three tients with endogenous depression were treated classes: 16 reported patient-level data; these data with nortriptyline 50 mg tid for 4 weeks.
sets are shown collectively in a subsequent section.
pression was rated on an idiosyncratic scale using They were also converted to aggregate data using scores 0–5. Concentration of nortriptyline in whole the thresholding scheme described in the methods blood was measured (although the article title men- section, and combined with the 7 studies reporting only aggregate data; these 23 studies entered into ported. Their averaged aggregate data do appear the meta-analysis. The 7 remaining studies reported in a figure in that paper suggesting that there may results without any supporting patient-level or ag- be a decline in response above 200 ng/mL. How- gregate data, and are described qualitatively in the ever, Fensbo concluded: “[N]o statistically signif- icant correlation between any blood concentration level and effect could be found. Moreover, neither ship between plasma concentration and improve- a lower or an upper bound limit for a therapeutic ment measured as fractional improvement (i.e., per- cent change) in the HAMD score. Using steady- Lipsey and colleagues [36] randomly assigned state values at 28 days, they found a therapeutic depressed post-stroke patients to nortriptyline (11 window of 40–107 ng/mL. Using data for trial com- completers) or placebo (15 completers) for 4– pleters at 10 weeks, they found a window of 42– weeks of treatment with a nortriptyline dose 111 ng/mL. No patients were treated above the up- titrated up to 100 mg at night. All patients who per limit of the consensus therapeutic window, 150 completed the trial were within the serum concen- tration range of 50–140 ng/mL by the end. The Thus, of the studies reporting only qualitative re- average HAMD score fell from 13.9 to 2.7; this sults, or aggregates not amenable to more detailed was statistically significant when compared to their analysis, three had negative results (that is, no evi- placebo group, but no patients were treated outside dence for the therapeutic window), three treated all patients within the therapeutic window (and thus Ng Ying Kin et al. [46] provided results of a trial were unable to disconfirm the window by testing (with additional details reported in [45]). In this higher concentrations), and one found a nonsignifi- trial geriatric patients (29 completers) were treated cant trend towards lower response at higher plasma with nortriptyline for up to 7 weeks.
was adjusted to maintain a serum level between 50 set of studies with widely disparate patient popula- and 170 ng/mL. They used a depression remission tions, trial designs, and treatment regimens is very threshold of a Hamilton depression rating score less than or equal to 10. Their conclusion about thera-peutic window was limited to the following state-ments: “The rate of clinical remission in this study Studies with quantitative patient-level
was relatively high compared to placebo. These re- sults tend to lend support for a similar therapeuticwindow in the depressed elderly.” Sixteen studies reported individual patient-level Bondareff et al. [6] described a trial involving out- data in tables or figures. All but one of these studies patients (70 completers) with DSM-III-R major de- used either the Cr ¨onholm-Ottosson depression rat- pressive disorder. They were treated with nortripty- ing scale or the Hamilton depression rating scale.
line for 12 weeks, with the dose adjusted to achieve For each of those two scales, composites of all data clinical response. They concluded: “The rate of re- using that scale are shown in Figure 1. One remain- sponse did not differ in patients with nortriptyline levels below 50 ng/mL or those with plasma lev- els 50–150 ng/mL, but patients with plasma levels At low plasma concentrations (less than 50 above 150 ng/mL showed suggestive but nonsignif- ng/mL), the antidepressant response increases as the plasma concentration increases, as would be ex- Streim et al. [59] reported on the treatment of When the plasma concentrations falls within the 69 frail, elderly, possibly demented, nursing home therapeutic window (50 to 150 ng/mL), the aver- residents who were randomized to treatment with age response (shown as the loess smoother) shows a a fixed dose after titration of either low dose or mildly inverted-U shape for the CO scale and a very normal dose of nortriptyline and treated for 10 slight increase with concentration for the HAMD For patients with relatively intact cogni- scale. The MADRS data set shows no clear trend tive function (Mini-Mental State Exam score > 18) (data not shown). However, focusing on the average [17], they found a quadratic (inverted-U) relation- response ignores the tremendous variation in an- tidepressant response. In general, under conditions of high variance, the average is not a very enlight- ening summary statistic. The distribution of the fi- nal depression scores is extremely broad, making it hard to use the window to guide treatment. Most importantly, many patients who were treated with concentrations in the window remained depressed.
When the patient level data are combined with the aggregate data, the positive predictive value is 52%, so knowing that the treatment is within the ther- apeutic window predicts a clinical response about half of the time. The corresponding positive predic- tive value for concentrations greater than the upperlimit of the window is 33%.
At plasma concentrations above the window (greater than 150 ng/mL) the response seems to plateau, and slightly declines. The possibility of aconfounding effect arising from dose escalation fornonresponsive patients will be explored below.
The GRADE quality of evidence for the patient- level data was assessed to be low because of the inconsistency and high variance of the results.
Meta-Analysis
The meta-analysis combined all studies for which patient-level or aggregate data were reported. The results are shown in Figure 2 for the default choices of the meta-analysis parameters: risk difference as the effect size measure, using the DerSimonian- Laird random effects method, a continuity correc- tion for all values, and comparing the window treat-ment to a control group, defined to contain those whose concentration is greater than the window up- The risk difference was calculated to be 0.17 (0.04–0.30, 95% CI), which is statistically significantat a 0.0105 level, rejecting the null hypothesis of no Figure 1: Composite of all data sets using Cr ¨onholm- effect. This risk difference corresponds to a number- Ottosson (upper) and Hamilton (lower) depression rating needed-to-treat of 5.9 (3.4–25, 95% CI). Incidental scales. Remission window (horizontal bar) and therapeu- tic window (vertical bar) are highlighted in gray. Loess nal publication promoting the idea of the nortripty- line therapeutic window, has the lowest risk differ-ence (smallest effect size) of all the studies exam-ined.
Figure 2: Forest plot for meta-analysis of treatment within nortriptyline therapeutic window versus treatment atconcentrations above the window. Effect size is computed as RD, the risk difference.
Heterogeneity is that part of the observed vari- and therefore small precision weights, they con- ation between studies due to true differences be- tribute relatively little to the summary effect. This tween them; the remainder of the variation is at- leads to wide confidence intervals, with most of tributed to random sampling. The I2 statistic re- the observed variation ascribed to random varia- ports the percentage of total variation not due to tion, not underlying heterogeneity. Given that these sampling. In this meta-analysis, I2 is 41.7, a level studies vary in many aspects of their design and of heterogeneity typically considered “moderate” implementation, it seems appropriate that the het- [22]. For the set of studies considered in this meta- analysis, one important factor to consider is that be-cause many of the studies have small sample sizes, Sensitivity analysis
Meta-Regression
To assess the effect of several of the study variables The meta-analysis was re-run using different pa- on the effect size, a meta-regression was performed, rameters. The choices include different methods using diagnosis (nonendogenous versus endoge- (fixed-effects or restricted maximum-likelihood esti- nous), design (observational versus clinical trial), mation (REML) instead of DerSimonian-Laird) dif- and dose regimen (not fixed versus fixed) as co- ferent effect size measures (odds ratio instead of risk variates. These were chosen because they seemed difference), a different treatment group (inside the a priori to be related to study quality; additional window) to those outside (either above or below) variables were not introduced because of the limited instead of only above, and various options for the amount of data available for the regression analysis.
continuity correction. Using no continuity correc- The meta-regression results appear in Table 2.
tion left only 11 studies, and was therefore omitted.
The coefficient for observational studies is largest There was no difference in using REML in the place in magnitude—this is due to the presence of the of DerSimonian-Laird, and little difference in using anomalous Montgomery study [41]—but falls short a fixed-effects model (results not shown). Different of statistical significance. The results for design and values of the continuity correlation and the control diagnosis are small and not statistically significant.
group also had little effect. The results are numeri-cally different when using an odds ratio effect size, Publication bias
but qualitatively the same, and similarly unchangedin corresponding sensitivity analyses.
Publication bias occurs when some results (usuallythose favoring the treatment) are preferentially pub- Studies varied quite a bit in their effect size val- lished. One method for detecting publication bias is ues and in their precisions. The influence of each through visual inspection of a funnel plot, which study was tested by leaving it out of the analysis in graphs the standard error against the risk differ- turn. Doing so reveals that the Montgomery study ence. Small studies will have large standard er- [41] had the most influence; it led to the lowest esti- rors; as the study size grows towards infinity the mate of the effect size when omitted, and produced sampling error approaches zero, so the funnel plot the lowest values for Q and I2. Its influence was should appear as a symmetrical funnel pointed up- also suggested by its outlying position in the forest wards. Publication bias is suspected when small plot. It produced an I2 of zero, meaning that the studies with negative results (here, lower values of observed variation of all the studies except for the the risk difference) are missing. The funnel plot for Montgomery study could be due to chance, with no the nortriptyline studies is shown in Figure 3.
difference in the true effect sizes. The Montgomerystudy is distinctive and not representative; some of Publication bias can be tested quantitatively us- the issues with this study were mentioned at the be- ing the trim and fill method, which imputes missing ginning of the results section. That the Montgomery studies and adds them to the funnel plot (shown in study occupies an outlying position is not to sug- the funnel plot as eight open circles). Both visual in- gest that the other studies are coherent evidence of spection and the trim and fill method suggest that a single, robust signal. It is more likely that many of studies may be missing from the lower right area of the studies are not precise enough to matter much the funnel, a region corresponding to studies with large standard errors (small sample size), with rel-atively large risk differences; that is, small studies The GRADE quality of evidence for the meta- showing response in the window. Such studies are analysis data was assessed as low, because of the unlikely to have been withheld from publication.
small study sizes, and the sensitivity analysis re- That is, in order for there to have been publica- tion bias, studies supporting the therapeutic win- data of limited evidential power as most or all pa-tients were treated within the therapeutic window,not allowing for disconfirmation of a declining re-sponse above the window’s upper limit.
Twenty-three studies had aggregate data that were used in a meta-analysis. Using commonly ac- cepted parameters in the meta-analysis produced a statistically significant effect for response within the window compared to response above the win- dow (risk difference of 0.17, p=0.01). This effect, as with that reported in the previous paragraph, was an aggregate one, based on disparate studiesof generally large variance and weak design. In the meta-regression, observational studies showed a stronger, but not statistically significant, response over clinical trials, but those studies were quite dis- parate and the strong response hinged on a singleoutlier. The effect of endogenous depression andfixed-dose regimen was small and not statisticallysignificant. A priori, one would expect that fixed- Figure 3: Funnel plot, using the trim and fill method to dose studies would be isolated from the confound- impute possibly missing studies (open circles); eight such ing effect of response on level, as clinicians might studies with a positive risk difference may be missing.
escalate doses for non-responders when allowed todo so; however, this effect was not detectable in the dow would have to have been suppressed, which limited data available. The evidence reported here seems improbable. Thus, no evidence of publication shows that even if a therapeutic window does exist, bias favoring the therapeutic window was found.
it was not reliably and robustly detected in studiesthat mimic typical clinical conditions, across typesof depression, and independent of the rating scale Discussion
used for measuring depression response.
The veridicality of the nortriptyline therapeutic This systematic review of all known published stud- window was an issue of some dispute in the 1970’s ies relating patient depression response to the nor- [18, 33, 2]. One prominent explanation for the dis- triptyline therapeutic window examined 16 data parity between those studies finding a window and sets reporting patient-level data, and found a slight those that did not was that the window response decline in response in those using the Cr ¨onholm- was primarily seen in endogenously depressed pa-Ottosson depression rating scale, starting within tients. As noted, the data reported here show a the window. In those using the Hamilton depres- trend towards this, but one not reaching statistical sion rating scale, response improved throughout the window, extended a bit beyond the window, then Overall, it does not appear that there is a clear showed a modest decline. These are group-level ef- role for clinical use of the therapeutic window. The fects; pronounced variation in response was seen predictive value of knowing that the patient’s con- both within and outside the therapeutic window.
centration is within the window is limited, and There is no evidence for a sharp cutoff at either the there is no strong cutoff at the consensus upper lower or upper window limits. Many of the studies limit of 150 ng/mL that would make it a useful were small and of low precision. Many provided target guiding prescribing practice. Independent of the therapeutic window, prudent practitioners vast majority of those in treatment.
would initiate treatment at lower doses, increase the The standard cautions about systematic reviews dose when indicated and as tolerated, assessing for and meta-analyses apply to this review. The results signs of response and side effects. It is not clear that depend on the completeness of the article search, the therapeutic window concept contributes much although. there was no evidence of publication bias beyond this procedure. As with other medications, that would have withheld negative results.
there remains value in the qualified use of plasma strengths of the surveyed primary studies are in the concentration monitoring for nortriptyline to ad- main quite limited. Many of the studies date from dress issues of patient adherence to recommended the 1970’s, a time when clinical studies were less rig- treatment, to investigate in cases where abnormal- orously designed, and standards for reporting were ities of metabolic function are suspected, and for lax. Many of the studies were small, one report- toxicologic purposes. Since nortripytline is metabo- ing relevant data on only 5 patients. Authors often lized into 10-hydroxy-nortriptyline, which appears assumed the window hypothesis to be correct, and to have antidepressant activity [5], a therapeutic did not design their studies to collect evidence that window measuring nortriptyline alone would be could disconfirm it by placing patients outside of compromised by the variability in the balance be- the window limits. In spite of the very limited evi- tween the competing pharmacokinetic and phar- dence, texts and reviews routinely cite the nortripty- macodynamic processes for those two compounds, line therapeutic window as if it were an established making the whole idea of a window for nortripty- Thus, weak, inconsistent evidence of highly vary- Previous reviews in this area have reached less ing quality for a nortriptyline therapeutic window nuanced conclusions. Typically they did not sur- was found. The inconsistency of the evidence war- vey all extant studies, or use quantitative meth- rants caution in generalization to patients outside of ods of evidence synthesis. Perry et al. [49] mod- the original treatment groups. Its clinical utility for eled the results of 7 studies using quadratic regres- prospective prediction of depression response over sion; the quadratic term was found to be not sta- routine, prudent prescribing practice appears to be tistically significant, although they argued for the existence of the window on other grounds. Perryet al. [50] used a receiver operating characteristic(ROC) analysis, finding optimal cutpoints for the Acknowledgements
therapeutic window of 58–148 ng/mL. However,the use of ROC curves in this manner requires that These efforts have spanned more than a decade. The costs be assigned to correct and incorrect classifi- cations, which, in their analysis, was implicit and peutic window was brought to my attention by Ian amounted to finding the location of crossing points Cook, MD, in a reading group during my psychi- of responder and nonresponder density functions, atry residency. UCLA Residency Training Director an arbitrary and poorly motivated choice. Ribeiro James Spar, MD, provided both funds to support et al. [53] conducted a more comprehensive meta- my encyclopedic photocopying efforts, and a very analysis; however, they missed two studies [47, 19], careful reading of the logic of the paper. Robert and produced a best estimate of the window us- V. Ashley, MD, Joel Braslow, MD, PhD, and Stephen ing an ROC analysis, again with the same failure to L. Read, MD, were enthusiastic supporters at vari- make explicit the costs of classification errors. They did not study the effect of study-level covariates us- Preliminary versions of a portion of this ma- ing a meta-regression. They reported a best-fit ther- terial were presented in the following two talks: apeutic window of 46–236 ng/mL, but this provides “What Evidence Supports the Nortriptyline Ther- little clinical guidance as it would encompass the apeutic Window,” Department of Veterans Affairs, Palo Alto Healthcare System MIRECC, Aug 26, 2009 [7] M. Borenstein, L. Hedges, J. Higgins, and and “Why We Think There is a Nortriptyline Ther- H. Rothstein. Introduction to Meta-analysis. Wi- apeutic Window: Clinical Epidemiology Meets the Sociology of Science,” Department of Veterans Af-fairs, Greater Los Angeles Healthcare System, West [8] G. Burrows, B. Davies, and B. Scoggins. Plasma LA VA Department of Psychiatry Grand Rounds, concentration of nortriptyline and clinical re- sponse in depressive illness. The Lancet, 300 22, 2009; participants at both sessions provided [9] G. Burrows, B. Scoggins, L. Turecek, and Version: this document produced August 5, 2013.
B. Davies. Plasma nortriptyline and clinical re-sponse. Clinical Pharmacology and Therapeutics,16(4):639–644, 1974.
References
[10] G. Burrows, L. Turecek, B. Davies, R. Mowbray, and B. Scoggins. A sequential trial comparing two plasma levels of nortriptyline. Australian and New Zealand Journal of Psychiatry, 8(1):21– [11] G. Burrows, K. Maguire, B. Scoggins, J. Steven- [2] A. Amdisen. Blood concentration of cyclic an- and clinical response—a study using changing tidepressants as a daily routine? A critical re- plasma levels. Psychological Medicine, 7(01):87– view from the doorstep of the clinical labora- tory. Acta Psychiatrica Scandinavica. Supplemen-tum, 61 Suppl 280:261–280, 1980.
[12] W. Cleveland and S. Devlin. Locally weighted regression: an approach to regression analysis Asberg, B. Cr ¨onholm, F. Sj ¨oqvist, and by local fitting. Journal of the American Statistical D. Tuck. Relationship between plasma level and therapeutic effect of nortriptyline. BritishMedical Journal, 3(5770):331–334, 1971.
[13] R. DerSimonian and N. Laird. Meta-analysis in clinical trials. Controlled Clinical Trials, 7(3): [4] P. Bech. The Cronholm-Ottosson depression scale: the first depression scale designed to [14] S. Duval and R. Tweedie. Trim and fill: a sim- ple funnel-plot-based method of testing and atrica Scandinavica, 84(5):439–445, 1991.
adjusting for publication bias in meta-analysis.
Pronounced inhibition of noradrenaline uptake [15] J. P. Feighner, E. Robins, S. B. Guze, R. A.
by 10-hydroxymetabolites of nortriptyline. Life Woodruff Jr, G. Winokur, and R. Munoz. Di- agnostic criteria for use in psychiatric research.
Archives of General Psychiatry, 26(1):57, 1972.
E. Richter, C. Clary, and E. Batzar. Comparison of sertraline and nortriptyline in the treatment and maprotiline with cardiovascular monitor- of major depressive disorder in late life. Amer- ing and serum level estimations. In A. Jukes, ican Journal of Psychiatry, 157(5):729–736, 2000.
editor, The Biochemical and Physiological Role of Ludiomil, pages 181–90. CIBA Laboratories: [26] L. Hollister, A. Pfefferbaum, and K. Davis.
Horsham, England, 1977.
Monitoring nortriptyline plasma concentra-tions. American Journal of Psychiatry, 137(4):485– [17] M. Folstein, S. Folstein, and P. McHugh. ‘Mini- mental state’. A practical method for gradingthe cognitive state of patients for the clinician.
[27] M. Keller. Remission versus response: the new J Psychiat Res, 12(3):189–198, 1975.
gold standard of antidepressant care. J Clin Psychiatry, 65 suppl 4:53–59, 2004.
D. Kupfer, R. Shader, J. Davis, C. B., J. Perel,G. Klerman, and D. Greenblatt. Tricyclic anti- depressants—blood level measurements and Hansen. Plasma-nortriptyline levels in endoge- clinical outcome: an APA task force report. Am nous depression. The Lancet, 301(7795):113–115, [19] M. Gold, A. Pottash, A. Stoll, D. Martin, [29] P. Kragh-Sørensen, C. Hansen, and M. ˚ L. Finn, and I. Extein. Nortriptyline plasma Plasma levels of nortriptyline in the treatment levels and clinical response in patients with fa- milial pure unipolar depression and blunted Scandinavica, 49(4):444–456, 1973.
trh tests. The International Journal of Psychiatryin Medicine, 13(3):215–220, 1983.
[30] P. Kragh-Sørensen, C. Eggert Hansen, P. Baas- trup, and E. Hvidberg. Self-inhibiting action [20] G. Guyatt, A. Oxman, G. Vist, R. Kunz, of nortriptyline’s antidepressive effect at high plasma levels. Psychopharmacology, 45(3):305– consensus on rating quality of evidence andstrength of recommendations. Bmj, 336(7650): [31] V. Kumar, R. Smith, K. Reed, and D. Leelavathi.
Plasma levels and effects of nortriptyline in geriatric depressed patients. Acta Psychiatrica Journal of Neurology, Neurosurgery, and Psychia- [32] L. Lehmann, C. Bowden, F. Redmond, and [22] J. Higgins, S. Thompson, J. Deeks, and D. Alt- B. Stanton. Amitriptyline and nortriptyline re- man. Measuring inconsistency in meta-analy- sponse profiles in unipolar depressed patients.
ses. Bmj, 327(7414):557–560, 2003.
Psychopharmacology, 77(2):193–197, 1982.
[23] L. Hollister. Monitoring tricyclic antidepres- [33] R. Levine. The role of plasma concentrations in sant plasma concentrations. JAMA: The Jour- the use of tricyclic antidepressant drugs. Prog nal of the American Medical Association, 241(23): Neuro-psychopharmacol, 3(1-3):211–222, 1979.
[34] A. Liberati, D. G. Altman, J. Tetzlaff, C. Mul- [24] L. Hollister. Plasma concentrations of tricyclic row, P. C. Gøtzsche, J. P. Ioannidis, M. Clarke, antidepressants in clinical practice. Journal of P. Devereaux, J. Kleijnen, and D. Moher. The Clinical Psychiatry, pages 66–69, 1982.
PRISMA statement for reporting systematic re- [25] L. Hollister, K. Davis, and P. Berger. Subtypes views and meta-analyses of studies that evalu- of depression based on excretion of MHPG and ate health care interventions: explanation and response to nortriptyline. Archives of General elaboration. Annals of Internal Medicine, 151(4): [35] M. Linder and P. Keck Jr. Standards of labora- [44] G. Murphy, A. Simons, and R. Wetzel. Plasma tory practice: antidepressant drug monitoring.
nortriptyline and clinical response in depres- Clinical Chemistry, 44(5):1073–1084, 1998.
sion. Journal of Affective Disorders, 8(2):123–129,1985.
[36] J. Lipsey, G. Pearlson, R. Robinson, K. Rao, and T. Price. Nortriptyline treatment of post-stroke [45] N. Nair, M. Amin, P. Holm, C. Katona, N. Kl- depression: a double-blind study. The Lancet, itgaard, N. Ng Ying Kin, P. Kragh-Sørensen, H. K ¨uhn, C. Leek, and K. Stage. Moclobemideand nortriptyline in elderly depressed pa- [37] W. Lyle, P. Brooks, D. Early, W. Leggett, G. Sil- tients. A randomized, multicentre trial against verman, R. Braithwaite, J. Cuthill, R. Goulding, placebo. Journal of Affective Disorders, 33(1):1–9, I. Pearson, R. Snaith, and G. Strang. Plasma concentration of nortriptyline as a guide totreatment. Postgraduate Medical Journal, [46] N. Ng Ying Kin, N. Klitgaard, N. Nair, M. Amin, P. Kragh-Sørensen, G. Schwartz, S. Ahmed, P. Holm, C. Katona, and K. Stage.
Clinical relevance of serum nortriptyline and 10-hydroxy-nortriptyline measurements in the treatment of depression with nortriptyline depressed elderly: A multicenter pharmacoki- netic and pharmacodynamic study. Neuropsy- hydroxymetabolites. Psychopharmacology, 92(2): [47] J. Pedersen and J. Sørensen. Therapeutic effect and side effects in patients with endogenous dence for the efficacy and safety of trazodone depression treated with oral nortriptyline once in insomnia. Journal of Clinical Psychiatry, a day. Neuropsychobiology, 6(1):42–47, 1980.
[48] P. Perry, J. Browne, B. Alexander, B. Pfohl, F. Dunner, A. Sherman, and M. Tsuang. Re- lationship of free nortriptyline levels to ther- sion scale designed to be sensitive to change.
apeutic response. Acta Psychiatr Scand, 72(2): Br J Psychiatry, 134(4):382–389, 1979.
[41] S. Montgomery, R. Braithwaite, and J. Cram- [49] P. Perry, B. Pfohl, and S. Holstad. The rela- mer. Routine nortriptyline levels in treatment tionship between antidepressant response and of depression. British Medical Journal, 2(6080): tricyclic antidepressant plasma concentrations.
A retrospective analysis of the literature using logistic regression analysis. Clinical Pharmacoki- 42] S. Montgomery, R. Braithwaite, S. Dawling, and R. McAuley. High plasma nortriptyline levels in the treatment of depression. I. Clin- [50] P. Perry, C. Zeilmann, and S. Arndt. Tricyclic ical Pharmacology and Therapeutics, 23(3):309– antidepressant concentrations in plasma: an estimate of their sensitivity and specificity asa predictor of response. Journal of Clinical Psy- [43] G. Murphy, A. Simons, R. Wetzel, and P. Lust- chopharmacology, 14(4):230–240, 1994.
man. Cognitive therapy and pharmacotherapy:Singly and together in the treatment of depres- [51] P. Perry, B. Alexander, L. B.I., and C. De- sion. Archives of General Psychiatry, 41(1):33–41, Vane. Psychotropic Drug Handbook. Lippincott Williams & Wilkins, eighth edition, 2006.
[52] R Development Core Team. R: A Language and Environment for Statistical Computing. R Foun- pression resistant to serotonin reuptake in- dation for Statistical Computing, Vienna, Aus- hibitors: case series. J Geriatr Psychiatry Neurol, tria, 2013. URL http://www.R-project.org.
[53] M. Ribeiro, E. Pereira, R. Santos-Jesus, E. Sena, [62] R. Young, G. Alexopoulos, R. Shindledecker, K. Petribu, and I. Oliveira. Nortriptyline blood levels and clinical outcome: meta-analysis of hydroxynortriptyline and therapeutic response published studies. Rev Bras Psiquiatr, 22(2):51– in geriatric depression. Neuropsychopharmacol- ogy: official publication of the American College ofNeuropsychopharmacology, 1(3):213–215, 1988.
[54] A. Rush, H. Kraemer, H. Sackeim, M. Fava, [63] V. Ziegler, P. Clayton, J. Taylor, B. Tee, and M. Trivedi, E. Frank, P. Ninan, M. Thase, A. Ge- J. Biggs. Nortriptyline plasma levels and thera- lenberg, D. Kupfer, et al. Report by the ACNP peutic response. Clinical Pharmacology and Ther- task force on response and remission in major depressive disorder. Neuropsychopharmacology,31(9):1841–1853, 2006.
[64] V. Ziegler, P. Clayton, and J. Biggs. A compar- ison study of amitriptyline and nortriptyline [55] A. Schatzberg and C. Nemeroff, editors. The with plasma levels. Archives of General Psychia- American Psychiatric Publishing Textbook of Psy- [56] R. Smith, K. Reed, and D. Leelavathi. Phar- macokinetics and the effects of nortriptyline ingeriatric depressed patients. Psychopharmacol-ogy bulletin, 16(3):54–57, 1980.
[57] B. Sorensen, P. Kragh-Sørensen, N. Larsen, and E. Hvidberg. The practical significance of nor-triptyline plasma control. Psychopharmacology,59:35–9, 1978.
[58] R. Spitzer, J. Endicott, and E. Robins.
search diagnostic criteria: rationale and relia-bility. Archives of General Psychiatry, 35(6):773–782, 1978.
[59] J. Streim, D. Oslin, I. Katz, B. Smith, S. Di- Filippo, T. Cooper, and T. Ten Have.
treatment of depression in frail elderly nurs-ing home residents. Am J Geriatric Psychiatry, 8(2):150–159, 2000.
[60] W. Viechtbauer. Conducting meta-analyses in R with the metafor package. Journal of Statisti-cal Software, 36:1–48, 2010.
Table 1: Studies reporting data on the nortriptyline therapeutic window. N is number of completers of treatmentwith nortriptyline at time closest to 4 weeks. Age is years, range or mean. NA=data not available. CT=clinicaltrial, Obs=observational study, Fixed=fixed dose of nortriptyline, Variable=variable dose, Clin=dose adjusted to clin-ical response, Level=dose chosen to achieve plasma level in range, CGI-I=Clinical Global Impression-Improvement,CO=Cr ¨onholm-Ottosson depression scale, HAMD=Hamilton depression rating scale.
Table 2: Model statistics for mixed-effects meta-regression using the variables diagnosis, design, and dose on theestimate of risk difference for treatment within the nortriptyline therapeutic window. Estimate is the estimate ofthe risk difference in the regression, SE is standard error, Z-value is the standardized estimate, CI is the confidenceinterval

Source: http://stevenbagley.net/blog/files/ntp.pdf