Efficacy of Antidepressants and USPSTF Guidelines for Depression Screening

The efforts of the U.S. Preventive Service TaskForce (USPSTF) to keep their evidence-based guidelines up to dateare praiseworthy. It is understandable that practical concerns maylimit speciﬁc questions addressed by new literature reviews. None-theless, it is disappointing that the questions addressed in the recentupdate of the depression screening guidelines (1) were not broader.In particular, the narrowly deﬁned review did not consider new ev-idence on the efﬁcacy of antidepressant treatment of depression from2 recent meta-analyses, which included many clinical trials whoseresults were not published and were heretofore inaccessible.Turner and colleagues (2) showed that results of 31% of com-pleted trials registered with the U.S. Food and Drug Administration(FDA) for 12 antidepressant drugs were provided to the FDA butnot published. Of these 23 unpublished trials, 16 did not showsigniﬁcant efﬁcacy per FDA review. Meta-analysis of all trials, pub-lished and unpublished, showed that antidepressants were less efﬁca-cious than in meta-analyses of only published trials. Furthermore,Kirsch and colleagues’ meta-analysis (3) of published and unpub-lished trials of 6 antidepressants, all of which provided results interms of the Hamilton Rating Scale for Depression, showed that thedrugs’ effects were barely clinically signiﬁcant. Thus, these 2 meta-analyses showed that failure to publish an important minority of“negative” trials of antidepressants biased the published literature infavor of these drugs, and that this bias made the drugs seem moreefﬁcacious than they really are. If these widely touted treatments ofdepression are really of marginal clinical effectiveness, the rationalefor screening for depression comes into question.Furthermore, the increasing number of cases in which clinicalresearch results that did not favor the interests of the commercialsponsors of the studies were suppressed suggests that we need torethink how evidence is gathered to support evidence-based practiceguidelines and perhaps how we regulate clinical research. A recenteditorial (4) in response to concerns about manipulating and sup-pressing evidence noted that “...thecurrent system isn’t working.Worse than that, it gives a false sense of security. The system’s fail-ures have left a legacy of drug evaluations for which, in the absenceof better information, we must assume...confusion and uncer-tainty...”.Failure to consider these issues when developingevidence-based guidelines just boots the problem down the road forothers to solve after it has become even more pernicious.


IN RESPONSE:
We agree that trade-offs may be involved in conducting targeted systematic reviews to efficiently update an evidencebased recommendation. We refer the reader to the USPSTF's procedure manual (1) for further clarification on their rationale and approach for conducting updates.
Dr. Poses suggests that explicit inclusion of unpublished trials in meta-analysis would reduce current estimates of antidepressant efficacy, thus undermining the rationale for depression screening in primary care. We do not believe an overestimate has been clearly established, and if it was, there are several reasons why it would not undermine the importance of screening. First and foremost, the rationale for screening relies most strongly on studies (key question 1) providing direct evidence that depression screening and care management programs increase depression response and remission compared with usual care. This evidence does not assume any particular efficacy of antidepressants per se, although efficacious treatments are part of the care pathway. Second, in the absence of direct evidence, the USPSTF's analytic framework for screening topics requires only that 1 or more primary care-applicable treatments be effective in screen-detected disease. For depression, pharmaceuticals are only one potentially effective treatment. The effectiveness of psychotherapy is not in dispute on the basis of unpublished studies. Thus, even if the overall efficacy of antidepressants were found to be less than estimated, this would not eliminate the potential value of screening.
Although we do not dispute the "file drawer problem" (publication bias in which null studies are less likely to be published), we believe that the studies by Turner and colleagues (2) and Kirsch and colleagues (3) do not in themselves refute the efficacy of antidepressants in major depressive disorder (the focus of our review). Turner and colleagues elegantly show not only publication bias but also distortion and inflation in effect sizes with their examination of published versus unpublished antidepressant trials provided to the FDA. Although they effectively showed reporting biases that would overestimate treatment efficacy, Turner and colleagues did not attempt a true meta-analysis of published and unpublished studies of antidepressant efficacy. The study by Kirsch and colleagues has several analytic problems limiting its conclusions, particularly the probable overestimate of the proportion of drug response attributable to placebo, due to use of between-study pooling rather than within-study pooling to gauge improvement.
Although we do not agree that our overall review findings about depression screening should be called into question, we appreciate Dr. Poses's emphasis that systematic reviews rigorously consider the full range of available data and potential biases involved in study reporting.

Smoking Cessation and the Risk for Type 2 Diabetes Mellitus
TO THE EDITOR: Yeh and colleagues (1) report that smoking cessation may increase the risk for type 2 diabetes mellitus. This may not be that surprising, given the well-known relationship between smoking cessation and weight gain, which has been exploited by cigarette manufacturers for many years (2). Yeh and colleagues (1) found that weight gain did, in fact, explain some of the excess risk for incident diabetes in smokers who quit. However, the mechanism of the increased risk for type 2 diabetes in these ex-smokers is undoubtedly more complex. Of note, smoking cessation can trigger a major depressive episode (3,4), which may put an individual at risk for type 2 diabetes as a result of being less attentive to healthy behaviors, such as following an optimum diet, taking prescribed medications, and exercising. Depression is also associated with decreased insulin sensitivity in persons at risk for type 2 diabetes (5). Furthermore, depression is associated with systemic inflammation (6), which may increase the risk for diabetes.
The article by Yeh and colleagues (1) should raise awareness that smoking cessation may increase the risk for diabetes. After individuals quit smoking, they may be more vulnerable to changes in mood that may have numerous effects on physical health and health risks. Although Yeh and colleagues did not report on depression in the patients in their study, the development of a major depressive episode should be considered as a possible mechanism for the important observation regarding smoking cessation as a risk factor for diabetes. TO THE EDITOR: Yeh and colleagues (1) prospectively examined the association of smoking status with development of incident type 2 diabetes in the ARIC (Atherosclerosis Risk in Communities) cohort (1) and concluded that predicts incident diabetes and that cessation leads to higher short-term risk. The first part of the conclusion was not convincing for several reasons.
First, excess risk did not emerge in one third of the smokers (mild smokers). Second, the study does not clarify whether this risk is valid only for men or for both sexes, because sex-stratified analyses are missing. In the regression analyses, the referent never-smokers and the top tertile comprised men and women who differed substantially in risk. Sex adjustment may not disclose the situation for women who may have been unaffected by the habit, although adjusted combined analyses have yielded excess risk. Third, adjustment for body mass index (BMI) may be biased depending on whether smoking led to a reduction in obesity in this cohort. If it did, which seems likely on the basis of the significantly lower BMIs in increasing categories of smoking, this would conceal a possible beneficial BMImediated effect of smoking on diabetes risk. This adjustment might introduce bias to male smokers as well. Finally, the adjusted hazard ratio was slightly increased at least a decade after smoking cessation and was increased 1.54-fold in ex-smokers who quit 3 to 5 years previously compared with never-smokers.
Although the introduction states that cigarette smoking is consistently identified as a risk factor for incident diabetes, the issue is far from established. Several studies (2,3) indicate that smoking may partly protect women against diabetes. In the ARIC cohort (4), persons with isolated impaired glucose tolerance, a prediabetic state in which people are more prone to diabetes, were less likely to smoke than those with normal or impaired fasting glucose. In most relevant studies, the group that developed diabetes had fewer smokers than the group that did not. DESIR (Data from an Epidemiological Study on the Insulin Resistance Syndrome) offers convincing findings to explain these observations. The authors were correct to consider that the effect of smoking on the proinflammatory state could mediate the development of diabetes; however, a large meta-analysis (5) in more than 160 000 participants noted equal or slightly lower C-reactive protein CRP concentrations in female current smokers compared with nonsmokers, although male current smokers had higher CRP levels. Thus, we would like to learn the effect of smoking on the risk for diabetes in each sex without mediation by BMI.

TO THE EDITOR:
We congratulate Yeh and colleagues (1) for a job well done. Tonstad and colleagues (2) had similar findings in patients with coronary heart disease: Ex-smokers were more likely to have the full metabolic syndrome and increased waist circumference. It is reported (3,4) that tobacco and other drugs of abuse are associated with appetite suppression and that discontinuation is associated with hyperphagia and weight gain. Discontinuation of drug use is so regularly associated with weight gain that most abstinenceoriented rehabilitation centers that provide detox services and treat addiction have incorporated prophylactic diet and exercise into their programs. Drugs of abuse and food compete at the same brain reinforcement sites; thus, cessation of an addictive substance can cause rebound hyperphagia and rapid weight gain, which is highly associated with the metabolic syndrome and diabetes (5).
It is unfortunate that data on eating habits were not available in the study by Yeh and colleagues, because frequency of consumption and types of foods consumed are important variables. Many addiction professionals and self-help groups teach that eating can soothe drug cravings and that avoiding hunger is a helpful tactic to keep from "falling off the wagon." Thus, some patients recovering from addiction carry sugar, and many recovery meetings have cakes, cookies, and candies readily available. Some research has found that sugar placebos have effects that augment recovery rates in smoking cessation. Research (6, 7) supports the idea that overconsumption of palatable foods (high fat and carbohydrate, high salt, or high carbohydrate) may be a sign of food addiction and another cause of obesity and subsequent issues.
Patients contemplating cessation will probably need a comprehensive yet individualized program that includes checking fasting blood glucose, fasting insulin (for example, to calculate quantitative insulin-sensitivity check), and CRP; behavioral and nutritional counseling; and, for some, pharmacotherapy, such as bupropion, to mitigate the orexigenic effects of nicotine withdrawal (8) to ensure that smoking cessation does not lead to overeating, weight gain, and increased risk for diabetes (9).
It is also possible that the systemic inflammation (another risk factor for diabetes) that is supposedly reduced during smoking cessation may increase because of rebound hyperphagia or a possible loss of ␤ 3 -receptor adipocyte influence on thermogenesis during nic-otine withdrawal, all resulting in rebound adiposity (10,11). More studies are needed to shed light on this issue.

IN RESPONSE:
We agree with Dr. Ziegelstein that smoking cessation may trigger depression, which could in turn increase the risk for type 2 diabetes because of unhealthy behaviors, decreased insulin sensitivity, or increased systemic inflammation. Unfortunately, the ARIC study did not include a specific assessment for depression. At the 3-year follow-up visit, ARIC participants were asked to complete a survey of vital exhaustion (1) to measure feelings of fatigue and exhaustion that might precede myocardial infarction. As Dr. Ziegelstein suspected, vital exhaustion was greater in quitters than in never-smokers (mean score adjusted for age, sex, race, and ARIC center, 11.3 [SD, 9.3] vs. 9.1 [SD, 8.3], respectively), but adjustment for vital exhaustion did not seem to explain the higher risk for incident diabetes in quitters in the fully adjusted regression model.
Dr. Onat makes several interesting comments. First, he is correct that smokers in the lowest tertile of cumulative pack-years had no excess risk for diabetes compared with never-smokers. The lowest-tertile group mostly comprised ex-smokers (77%) who had probably quit more than 10 to 15 years before enrollment in ARIC. Therefore, the results seem consistent with our finding that diabetes risk decreases to the background rate during this time. Second, he suggests that the common practice of introducing BMI into multivariate models that relate smoking to diabetes risk might constitute inappropriate overadjustment for a risk mediator (insofar as smoking-related weight loss might contribute to reduced diabetes risk) that would bias risk estimates upward. We therefore did an additional analysis excluding BMI and waist circumference from the fully adjusted model. This led to a small attenuation in the risk estimates, but the relationship remained graded and statistically significant. Finally, Dr. Onat questions the appropriateness of pooling men and women into a single analysis, because the effects of smoking on weight might differ by sex. We therefore assessed for possible interactions in models with and without BMI and waist circumference. There was no evidence of effect modification by sex in either model. We therefore stand by our pooled approach.
Finally, the comments by Dr. Shriner and colleagues regarding potential changes in eating habits after quitting are well taken. Unfortunately, we cannot test their hypothesis in the ARIC cohort. The ARIC study used a food-frequency questionnaire to collect usual dietary intake, which does not reflect daily eating behavior, but rather long-term dietary patterns. Dr. Shriner and colleagues also recommend checking fasting insulin and CRP levels as part of a postcessation monitoring plan. Although additional assays would provide further information, demographic and clinical factors readily available in the primary care setting without extra cost seem to carry a great deal of predictive information: Older men who were heavy smokers and who gained more weight after quitting have by far the highest risk for diabetes. Physicians and patients should be aware of those risk factors even without additional blood assays.

Opioids in Management of Chronic Noncancer Pain
This conversation continues on the Web. To read more, go to http: //annals.org/content/152/2/123.full/reply#annintmed_el_124279.

TO THE EDITOR:
In addition to the overdose and mortality risks from prescription opioids discussed by McLellan and Turner (1), 3 additional opioid-related issues require attention: the possibility that many persons who have overdosed or died of illegally diverted prescription opioids (or heroin) initially became addicted through the use of a legitimate prescription; the prominent role of psychological disorders, which have substantially contributed to the overprescription of opioid analgesics; and the likelihood that extendedrelease and long-acting opioids carry greater risks for addiction, overdose, and death than shorter-acting agents.
Many persons who overdosed or died of diverted prescription opioids or heroin (obtained either from the street or from friends or relatives) may have initially been prescribed an opioid for some type of pain, which led to addiction and then to the drug-seeking behaviors associated with this disease. Likewise, many addicted persons who obtain most of their opioids through illegal "doctor shopping" may have developed their addiction through an initial or continued legal prescription for pain medicines. These common situations underscore the importance of prudent, selective prescribing of opioids for specific clinical indications for chronic noncancer pain, with close attention to the risk-benefit ratio and to recently established guidelines, as well as to careful monitoring and knowing when to discontinue or taper these potent brain-active drugs.
As stated by McLellan and Turner, opioid overdose and death were associated with depression, benzodiazepine use, and history of substance abuse. This suggests that many opioid prescriptions are written for chronic noncancer pain in which psychological disorders, such as anxiety syndromes and depressive disorders, are present but often missed. Because psychological comorbid conditions are common in chronic noncancer pain disorders, particularly fibromyalgia (2, 3), treatment with nonopioid agents and nonpharmacologic approaches is indicated, possibly including referral for psychological or behavioral therapies (4).
My observation from clinical practice and as a national disability peer-reviewer in rheumatology is that many patients have undiagnosed or underdiagnosed psychological disorders that have been inadequately addressed by focusing mainly on chronic pain misattributed solely to a somatic structure, such as degenerative discs, arthritis, or muscle tissue (such as in fibromyalgia). These somatic labels then become the indication to prescribe opioids for presumed tissue pain, while the psychological disorders underlying the pain go unrecognized. This predisposes to persistence of symptoms, drug dependence, and continued dysfunction and increases claims for long-term disability. These factors have become a major contributor to the rising health care costs affecting our nation and, unless recognized, will continue to fuel the prescription opioid epidemic and its adverse consequences, including addiction, overdose, and death.
Finally, as noted by the authors, the study by Dunn and colleagues (5) focused on short-acting opioids used by more than 90% of the cohort, but if extended-release and long-acting opioids, such as oxycontin and methadone, had been the main drugs studied, the outcomes would most likely have been worse, especially because oxycodone and methadone have resulted in the highest reported mortality rates (6). We hope to see further studies and discussions about the problems of prescription opioids, which continue to take a large toll on the public health and social systems of our nation.  (1), written in response to the study by Dunn and colleagues (2). As an addiction medicine specialist and an American Board of Medical Specialties-certified pain specialist, my teaching often focuses on the substance abuse and psychosocial aspects of the therapeutic use of controlled substances.

Stephen G. Gelfand, MD
I will not comment on the cohort study on opioid overdose or the general details of the editorial itself. Both publications make important contributions to the field of pain medicine and public health, and I predict that the issues raised will stimulate thinking and responses that will lead to important academic and policy dialogue. However, the authors should consider modifying the statement that "[p]rescribing opioids at high doses is both dangerous and questionable for indications other than methadone treatment of opioid dependence," by substituting "can be" for "is." Those familiar with the literature on pain medicine (3-6) and the standards of care across the United States (7) know that the statement overreaches and seems to overlook the ostensibly few cases of both malignant and nonmalignant pain in which the appropriate use of high-dose opioids has been both successful and accepted by the medical community and has become an essential part of the practice of medicine for a particular patient. Both the literature and the medical board guidelines emphasize diligent selectivity and continuous and careful monitoring, particularly when the opioid dose is substantial and long-term, but wisely avoid absolute statements that question or label as dangerous a particular clinical practice.
Because this statement has the imprimatur of both the White House Office of National Drug Control Policy and Annals, and because adverse consequences may occur both therapeutically and in medicolegal contexts as a result of this statement, I believe it is important to correct it.  (1) states that "Opioid overdose is among the most common causes of death nationwide." Reference 3 gave a total of 9798 deaths from opioids plus cocaine in 2004. Because some deaths were due to cocaine, the number of opioid deaths was less than 9798. They may have meant the most common cause of accidental deaths or poisoning deaths, but deaths from opioid overdose are certainly not more common than deaths from cancer, heart disease, or stroke.

Marcus M. Reidenberg, MD Weill Cornell Medical College New York, NY 10065
Potential Conflicts of Interest: None disclosed.

IN RESPONSE:
The remarks from Dr. Gelfand and Dr. Kornfeld exemplify the wide spectrum of opinions about the use of opioids for chronic noncancer pain. Dr. Gelfand questions the use of these drugs in persons with mental health disorders, and Dr. Kornfeld takes issue with our suggestion that high-dose opioids should be avoided. However, Dr. Gelfand accurately emphasizes the high prevalence of comorbid mental health disorders in persons prescribed opioids. In a national population-based community survey (1), persons with a history of mood disorder (such as depression and anxiety disorder) were more than 4 times more likely than persons without such a history to report subsequent opioid treatment. We agree that physicians need to address mental health disorders when treating pain, but we would not deny a potentially beneficial therapy to patients just because they have such a disorder.
In his remarks supporting the use of high-dose opioids, Dr. Kornfeld does not distinguish whether this treatment is for cancer or noncancer pain, but our editorial and the article by Dunn and colleagues (2) concern only noncancer pain. The American Pain Society-American Academy of Pain Medicine Opioids Guidelines Panel stated that " . . . there is little evidence to guide safe and effective prescribing at higher [opioid] doses and there is no standardized definition for what constitutes a 'high' dose" (3). The Panel raises serious concerns, as do we, about high-dose opioid therapy (which they define as oral morphine or equivalent, Ͼ200 mg/d) because of the risks for diversion, opioid-related adverse effects, changes in health status, and poor adherence to the opioid treatment plan.

www.annals.org
If patients require high-dose therapy, the panel recommends using complementary nonopioid therapies to help reduce reliance on opioids and possibly rotating opioids along with dose reductions (3). An American Pain Society-American Academy of Pain Medicine panel identified related research gaps, including, "What are the benefits and harms of high (Ͼ200 mg/d of morphine or equivalent) versus lower doses of opioids for chronic noncancer pain?" and "Are high doses of opioids associated with different or unique harms compared with lower doses?" (4). Such fundamental gaps in our knowledge, along with worrisome evidence of serious risks, raise serious questions about the use of high-dose opioids to manage noncancer pain.
Finally, we thank Dr. Reidenberg for his correction. The sentence in our editorial to which he refers should have read, "Opioid overdose is among the most common causes of accidental death nationwide." This has been corrected in the online version.

Bilateral Necrosis of Earlobes and Cheeks: Another Complication of Cocaine Contaminated With Levamisole
Background: Levamisole is a veterinary antihelmintic agent that has also been used as an immunomodulating agent to treat cancer, autoimmune diseases, and the nephrotic syndrome. It has been shown to cause neutropenia and vasculitis (1)(2)(3). The practice of cutting cocaine with levamisole has been increasing since 2005, and in 2009, 70% of cocaine seized at U.S. borders contained levamisole (4). Recent reports (4, 5) have implicated levamisole as a cause of neutropenia in cocaine abusers. Two cocaine abusers with similar cases of neutropenia and vasculitis presented to the University of Rochester Medical Center, New York, within 8 days of each other. We hypothesize that cocaine contaminated with levamisole may cause not only neutropenia but also vasculitis.
Case Report: The first patient was a 57-year-old woman who presented with 2 days of worsening cutaneous lesions, fevers, chills, arthralgias, and flare-ups of recurrent Staphylococcus aureus boils. She had no significant medical history except for smoking cocaine, with her last use 4 days before presentation. On examination, she had tender palpable purpuric plaques on both cheeks and ear lobes with central areas of necrosis (Figure).
On the day of admission, the patient's absolute neutrophil count was 2.3 ϫ 10 9 cells/L; within 48 hours, it dropped to 0.5 ϫ 10 9 cells/L. Anticardiolipin IgM levels were low-to mediumpositive. Perinuclear antineutrophil cytoplasmic antibody (P-ANCA) was positive (titer Ͼ 1:2560). Further work-up for hematologic and rheumatologic disorders was negative. Histopathologic examination of her right cheek lesion showed organizing thrombi in small-caliber vessels associated with perivascular lymphocytic inflammatory cell infiltrates; these findings were consistent with a thrombotic disorder and would be less likely with small-vessel vasculitis. The skin lesions resolved spontaneously, and neutropenia was corrected by treatment with filgrastim.
The second patient was a 22-year-old woman who presented with 2 weeks of a painful rash on her face, ears, legs, thighs, and buttocks. Her medical history was significant for cocaine abuse. Examination revealed many exquisitely tender erythematous plaques with central necrosis on her thighs, lower extremities, and buttocks. Her cheeks had erythematous plaques, and her nose had a prominent necrotic lesion on an erythematous base.
Laboratory data showed an absolute neutrophil count less than 1.0 ϫ 10 9 cells/L. Her P-ANCA results were positive (titer, 1:2560). Anticardiolipin IgM antibody and plasma serine IgM levels were elevated. Complement and protein C and S levels were moderately low. Results of further work-up for hematologic and rheumatologic disorders were negative. Toxicology screening was positive for cocaine.
A skin biopsy showed small-vessel vasculitis with associated multiple intravascular thrombi consistent with leukocytoclastic vasculitis. The patient was treated with steroids, and her lesions and neutropenia improved rapidly.
Discussion: Both patients presented with a clinical, histologic, and immunologic profile typical of levamisole toxicity. Although levamisole levels were not obtained, exposure is very likely because 70% of cocaine is contaminated by levamisole (4). Levamisole has serious side effects, such as bone-marrow suppression; skin reactions, including cutaneous vasculitis, are less frequent. Vasculitis has a very distinct presentation, with purpuric lesions preferentially involving the ear lobes and an association with anticardiolipin and P-ANCA levels similar to the presentation of our patients. Histologic examination of the skin lesions shows a vasculopathic reaction pattern ranging from leukocytoclastic and thrombotic vasculitis to vascular occlusive disease without true vasculitis (2,3). Spontaneous resolution occurs by discontinuing levamisole use.
Conclusion: Clinicians should suspect exposure to cocaine contaminated with levamisole in patients presenting with both neutropenia and necrotic skin lesions. Testing for levamisole in urine and