Effects of Supraphysiologic Doses of Testosterone on Mood and Aggression in Normal Men: A Randomized Controlled Trial

Background  Field studies of illicit anabolic-androgenic steroid users suggest that some develop manic or aggressive reactions to these drugs—a potential public health problem. However, controlled laboratory evaluations of these effects remain limited.

Methods  In a randomized, placebo-controlled, crossover trial, we administered testosterone cypionate for 6 weeks in doses rising to 600 mg/wk and placebo for 6 weeks, separated by 6 weeks of no treatment, to 56 men aged 20 to 50 years. Psychiatric outcome measures included the Young Mania Rating Scale (YMRS), the Point Subtraction Aggression Paradigm (a computerized provocation test of aggression), the Aggression Questionnaire of Buss and Perry, the Symptom Checklist-90-R, daily diaries of manic and depressive symptoms, and similar weekly diaries completed by a "significant other" who knew the participant well.

Results  Testosterone treatment significantly increased manic scores on the YMRS (P=.002), manic scores on daily diaries (P=.003), visual analog ratings of liking the drug effect (P=.008), and aggressive responses on the Point Subtraction Aggression Paradigm (P=.03). Drug response was highly variable: of 50 participants who received 600 mg/wk of testosterone cypionate, 42 (84%) exhibited minimal psychiatric effects (maximum YMRS score, <10), 6 (12%) became mildly hypomanic (YMRS score, 10-19), and 2 (4%) became markedly hypomanic (YMRS score, ≥20). The 8 "responders" and 42 "nonresponders" did not differ significantly on baseline demographic, psychological, laboratory, or physiological measures.

Conclusions  Testosterone administration, 600 mg/wk increased ratings of manic symptoms in normal men. This effect, however, was not uniform across individuals; most showed little psychological change, whereas a few developed prominent effects. The mechanism of these variable reactions remains unclear.

THE MALE hormone testosterone and its synthetic analogs compose the family of hormones called anabolic-androgenic steroids (AASs). Probably more than a million Americans, primarily young men—including 4% to 7% of male high school students—have used these hormones illicitly to improve athletic performance or personal appearance.^1-4 During the past 10 years, results of a growing literature of field studies^1,5-13 suggest that some illicit AAS users develop marked aggression, hypomania, and occasionally frank mania during AAS exposure, as well as depressive symptoms and even suicidality during AAS withdrawal. The euphoric effects of AAS use and the dysphoric effects of withdrawal may contribute to a syndrome of AAS dependence in some individuals.^1,13-15 Several articles^1,8,16-22 speculate that AAS use may precipitate criminal violence. These effects may be dose related, with frequent symptoms in individuals using the equivalent of more than 1000 mg of testosterone per week,^5-7 occasional symptoms at intermediate dosages,^6,8-13 and few symptoms at 300 mg per week or less.^6,23 In one field study of 88 illicit AAS users,⁶ 25 (28%) reported using at least 1000 mg of testosterone or the equivalent per week, and another 51 (58%) had used between 300 and 1000 mg per week—often as "stacks" of oral and injectable AAS taken simultaneously. Thus, many illicit users may be at risk for psychiatric morbidity.

However, interpretation of these findings is limited by the problems common to observational studies. For example, selection bias may affect which users will present for study, and information bias may arise in participants recalling psychological effects experienced in the past while taking illicit drugs of uncertain potency or authenticity. The confounding effects of users' premorbid personalities and expectations, concomitant abuse of other substances, and the physiological and psychological effects of weight training might also bias the findings.^6,24-26

Experimental treatment studies avoid such limitations but ethically cannot use highly supraphysiologic doses of AAS approaching those used illicitly. Endocrinologic,^27,28 physiological,^29-31 and medical^32-34 studies have typically used at most 300 mg/wk of testosterone^27-34 or nandrolone²⁹ and have noted few psychiatric effects. However, 3 recent laboratory studies have used higher doses: 2 reported occasional manic or hypomanic reactions in participants administered methyltestosterone³⁵ or testosterone,³⁶ whereas a third study using testosterone did not.^37,38

To augment these limited data, we performed a randomized, placebo-controlled, double-blind crossover study of the psychiatric effects of intramuscular testosterone cypionate treatment, at dosages rising to 600 mg/wk, in 56 normal men. We hypothesized that this supraphysiologic dosage of an AAS would increase hypomanic and aggressive symptoms. We also tested the hypothesis that such effects, if present, would be more prominent in men who had previously used AASs illicitly or who lifted weights regularly.

We advertised at several local colleges and 1 local gymnasium to recruit men aged 20 to 50 years from 3 groups: (1) men who did not lift weights regularly and never used AASs, (2) men who lifted weights regularly (at least 3 times per week for ≥2 years) and never used AASs, and (3) men who formerly used AASs illicitly. We required each participant to furnish a "significant other"—a spouse, sexual partner, or close friend—to rate his behavior in a weekly diary (see the "Study Procedures" subsection) and to alert the investigators if he displayed adverse behavioral changes. We informed each participant and significant other that we were investigating the psychological effects of receiving supraphysiologic doses of testosterone, and required both to sign informed consent forms approved by the institutional review board of McLean Hospital, Belmont, Mass.

At baseline evaluation, we obtained demographic information, psychiatric and substance abuse histories using the Axis I portion of the Structured Clinical Interview for DSM-III-R,³⁹ and medical histories. The evaluation also included a physical examination, determination of body fat using calipers,⁴⁰ an electrocardiogram, and laboratory tests (see the "Study Procedures" subsection). We excluded participants who (1) met DSM-III-R criteria⁴¹ for any substance abuse or dependence within the past year; (2) reported current or past major depression, active suicidal ideation, hypomania, mania, or psychotic symptoms; (3) had ever required use of any regular psychiatric medication; (4) exhibited a clinically significant medical condition; or (5) had used any AASs within the past 90 days, as determined by self-report and baseline urine testing.

Qualifying participants were then seen weekly for 25 weeks. They were randomized to receive intragluteal injections of testosterone or placebo under double-blind conditions forweeks 1 through 6 (the "first treatment period"), a washout for weeks 7 through 12, injections of the opposite treatment for weeks 13 through 18 (the "second treatment period"), and a second washout for weeks 19 through 25. During the testosterone treatment period, participants received 150 mg each of the first 2 weeks, 300 mg the third and fourth weeks, and 600 mg each of the last 2 weeks. Placebo injections consisted of equivalent amounts of the sesame oil vehicle. We assessed psychiatric symptoms at each visit using the Young Mania Rating Scale (YMRS),⁴² 24-item Hamilton Depression Rating Scale,⁴³ Aggression Questionnaire of Buss and Perry (AQ),⁴⁴ and Symptom Checklist-90-R.⁴⁵ Weight, pulse, and blood pressure were also measured weekly. Laboratory tests—including standard chemistries, hematological measures, neuroendocrine measures (total testosterone, luteinizing hormone, follicle-stimulating hormone, and prolactin), urinalysis, and urine screening for drugs of abuse (including AASs^46,47)—were administered every 3 weeks. Blood and urine samples for these tests were obtained within a few minutes before or after any intragluteal injections. To minimize the possible confounding effects of diet and weight training,²⁴ we asked participants to maintain a stable diet and exercise pattern throughout the 25 weeks, although this behavior was not formally monitored. Virtually all injections and investigator-administered ratings were performed by one of us (H.G.P.); in his absence, they were performed by another physician (J.I.H.). Thus, the raters were aware of when participants were receiving treatment but were unaware of the identity of that treatment.

At home, participants also completed a 17-item daily diary (available on request) covering 12 manic symptoms and 5 depressive symptoms experienced during the past 24 hours. The diaries generated a manic score of 0 to 48 and a depression score of 0 to 20. Participants also marked a 100-point visual analog "liking" scale with 3 anchors: "I dislike the way that I feel on this medication very much" (0 points), "neutral" (50 points), and "I like the way that I feel on this medication very much" (100 points). Total manic, depressive, and liking scores for each week represented the mean scores of all diaries submitted during the preceding 7 days. Significant others rated the participant's behavior weekly in a virtually identical diary. Participants received approximately $50 per week for study participation and significant others received $5 per weekly diary.

In the second study year, we introduced a computerized measure of aggression, the Point Subtraction Aggression Paradigm (PSAP).⁴⁸ This test was administered to most participants at weeks 1, 5, 7, 13, 17, and 19. We presented this technique in detail in a previous article,⁴⁹ together with preliminary data from 5 participants in the present study. Briefly, each participant was seated in a booth equipped with a monitor screen and was told that he was playing against an unseen male opponent, which was actually a computer. The participant could accumulate points on the screen—exchangeable for money—by pressing one button, or he could deprive his "opponent" of points by pressing another button. During the session, the opponent provoked the participant by randomly depriving him of points. The participant's aggression score represented the total number of points that he subtracted from his opponent in retaliation to this provocation. In accordance with published scoring methods for this test,⁴⁸ we excluded participants who guessed that their opponent was a computer. Participants were judged to have guessed correctly if they both (1) failed to deprive their opponent of points during 1 or more sessions, and (2) responded on a poststudy questionnaire⁴⁹ that their opponent had been a computer.

For the primary analysis of each outcome measure, we compared the changes during the testosterone and placebo periods. We defined each period to begin on the day of the first injection and end 1 week after the last injection (when the effects of testosterone would be expected to be greatest). We considered participants to be "fully evaluable" for a given treatment period if they received at least the first 5 of 6 scheduled injections to ensure that all testosterone treatment periods included at least 1 injection of 600 mg of testosterone. In addition to the analysis of fully evaluable participants, we also analyzed participants' responses after adminstration of 300 mg of testosterone, using as the end point the visit 1 week after the second injection of 300 mg (week 5 or week 17).

We used a model for the mean of the response variable derived from Senn⁵⁰ to analyze crossover studies. This model incorporates terms for treatment and period effects but not for carryover (or residual) effect. The decision to exclude such a term was based on the work of Freeman,⁵¹ who showed that models with a term for carryover effect are biased and possess poor power to detect a carryover effect. In any event, a carryover effect seemed unlikely with our design because evidence from kinetic studies^52,53 with testosterone cypionate suggests that a 6-week washout period after a 6-week treatment period allows neuroendocrine function to return to baseline. Thus, the model was as follows: E(Y_i)= β₀ +β₁ treatment + β₂ period, where E(Y) is the expected value of the outcome variable for the ith subject; treatment is an indicator variable for testosterone treatment; and period is an indicator variable for the first treatment period.

To model the covariance, we wanted to allow for both dependence of repeated observations within the same individual and possible dependence of variance of the response variable on the mean. Therefore, we used PROC MIXED in SAS⁵⁴ to fit 3 models for the "working" covariance: one allowed variance to depend on treatment group, another allowed variance to depend on treatment period, and a third assumed equal variance (compound symmetry). We chose the best-fitting model in each analysis and computed SEs based on empirical variances.

We tested for effects of previous AAS use and/or regular weight lifting by putting terms in the model for the various participant categories (1 indicates no weight lifting, no previous AAS use; 2, regular weight lifting, no previous AAS use; and 3, previous AAS use). Because it was unclear whether this variable should be best treated as a nominal or an ordinal variable, we analyzed it both ways.

We also analyzed data from only the first treatment period using the Wilcoxon rank sum test, 2-tailed. This analysis was free of concern about carryover effect and distributional assumptions but ignored almost half of the data. Thus, it represented an unquestionably valid but overly conservative assessment of treatment effects.

In an a posteriori analysis, we classified participants' responses to testosterone treatment as marked, moderate, or minimal on the basis of YMRS scores (see "Results" section), and then compared attributes of marked or moderate responders with those of minimal responders using the Fisher exact test, 2-tailed. In these and all other comparisons, we considered P<.05 to be statistically significant.

Because the outcome measures in this study are correlated to an unknown degree, it is difficult to calculate an appropriate correction for the effect of multiple comparisons. Results are therefore presented without correction. Thus, some findings, especially those of marginal significance and those that do not test a priori hypotheses, may represent chance associations.

Of 66 participants recruited, 56 were randomized, 55 were followed up in the first study period; 53 received at least 5 injections and were considered fully evaluable, as defined in the "Data Analysis" subsection of the "Participants and Methods" section (demographic characteristics of these 53 men are shown in Table 1). Forty-nine participants were followed up in the second study period; 47 were fully evaluable.

Manic scores on the YMRS increased significantly with testosterone treatment (Table 2). However, this effect was not uniform: most participants exhibited little change during testosterone treatment, whereas a few displayed marked symptoms (see Figure 1 and the "Comment" section). While taking testosterone, no participant reported actual violence (see definition in Table 1), but several described instances of uncharacteristic aggressiveness; we withdrew one participant from the study after the fifth week because he became alarmingly hypomanic and aggressive (Figure 1). Of 34 participants administered the PSAP at baseline and after at least 1 treatment period, 1 was excluded because he misinterpreted the instructions and 6 because they guessed that their opponent was a computer (see "Study Procedures" subsection of the "Participants and Methods" section). Thus, 27 participants were evaluable for at least 1 study period. Because some participants were not given the PSAP at week 13 (the baseline for the second period), we analyzed this measure using the end-point value (week 7 or week 19) for testosterone vs placebo treatment, with the week 1 value as a covariate. Despite the smaller number of participants in this analysis, aggression scores on the PSAP increased significantly with testosterone treatment. Again, however, this effect was not uniform; most participants showed little change and a few showed marked changes. The manic and liking scores on participants' daily diaries also showed a significant testosterone treatment effect. Diaries from significant others did not show a significant testosterone effect, although manic scores on participants' and significant others' diaries correlated highly at the end of the testosterone treatment period (Spearman Y=0.59; P<.001).

Self-rating scales showed few drug treatment effects. Only 1 of 4 AQ subscale scores, verbal hostility, increased significantly with testosterone therapy among evaluable participants (P=.03), and this score displayed a significant period effect, with second-period increases greater than those of the first period (P=.01). Only 1 of 9 Symptom Checklist-90-R subscale scores, phobic anxiety, showed a significant increase with testosterone treatment (P=.006). Depression scores on both types of diaries and on the Hamilton Depression Rating Scale remained low, with no changes even approaching significance during testosterone administration or withdrawal. When we expanded our analysis to include the 55 participants who had received at least 1 injection, the effect of testosterone treatment and the levels of significance on all primary psychiatric measures remained virtually unchanged.

We examined whether the categories of weight lifting and previous AAS use (no regular weight lifting and no previous use, regular weight lifting without previous use, previous use, and both weight lifting and previous use groups combined) were associated with a significant effect on outcome measures. We found no significant effect or even tendency toward an effect (P<.10) on any outcome measure in Table 2, regardless of whether these categories were treated as ordered or unordered.

Next, by examining response to testosterone treatment at the 300-mg level, as discussed in the "Participants and Methods" section, we found only a slight effect. The effect of testosterone treatment (mean±SE) was 1.1±0.6 (χ²₁=3.68; P=.06) on the YMRS, 1.2±0.6 (χ²₁=4.20; P=.04) on the daily diary manic score, and 3.4±0.1 (χ²₁=4.49; P=.03) on the daily diary liking score. The effect did not approach significance on the PSAP at 47±39 (χ²₁=1.41; P=.23), or on the other measures.

Looking at the first treatment period only, the changes on psychiatric measures were similar to those found in the primary analysis but yielded lower levels of statistical significance. Of the 4 measures showing a significant effect for testosterone treatment in the primary analysis, the change in YMRS scores was significantly greater with testosterone than with placebo treatment (median, 2 vs 0; z=2.96; P=.003). All changes on the other 3 measures showed a tendency toward statistical significance in the same direction (PSAP: median, 99 vs 14; z=1.71; P=.09; daily diary manic scores: median, 0.1 vs -1.0; z=1.66; P=.10; liking scores: median, 0 vs 0; z=1.62; P=.11). As with the primary analysis, the difference between testosterone and placebo treatment on the other 4 primary psychological measures was not significant.

Finally, in an a posteriori analysis, we divided the evaluable participants into 3 groups based on maximum YMRS score attained during each treatment: marked responders (YMRS score of ≥20, indicating manic symptoms likely to impair social and occupational functioning), moderate responders (YMRS scores of 10-19, indicating milder hypomanic symptoms), and minimal responders (YMRS scores of <10). Because there were no significant period effects in the analyses of YMRS data, we scored participants' responses to testosterone treatment, regardless of the period during which it was administered. This classification yielded 2 marked, 6 moderate, and 42 minimal responders to testosterone treatment (Figure 1). By comparison, the 50 placebo periods produced no marked, 1 moderate, and 49 minimal responses. Diary scores seemed to be consistent with this classification: manic scores of the 8 moderate or marked responders increased significantly more than those of the 42 minimal responders (median, 6.9 vs 0.2; z=3.90; P<.001 by Wilcoxon rank sum test, 2-tailed), as did liking scores (median, 6 vs 0; z=2.93; P=.003) and manic scores on the significant other diaries (median, 8.9 vs -0.2; z=3.48; P<.001). We then compared the 8 moderate or marked responders with the 42 minimal responders on all baseline variables in Table 1 and on all physiological and laboratory variables in Table 3. The latter were examined at baseline and at the end of the testosterone period. No comparison yielded a significant difference. Also, none of the 11 individuals reporting a first-degree relative with a major mood disorder were among the 8 testosterone responders.

Although diet, exercise, and times of blood drawing were not standardized across participants in this primarily psychiatric study, testosterone treatment produced expected changes on physiological and laboratory test results (Table 3).^27,37,56 Measures differing significantly between testosterone and placebo treatment among the 53 fully evaluable participants are summarized in Table 3; all other measures on the standard chemistry, hematologic, and urinalysis battery produced no significant differences. No serious laboratory abnormalities or adverse medical events occurred during the study.

In a placebo-controlled, double-blind, crossover study of administration of supraphysiologic doses of testosterone cypionate to normal men, testosterone treatment significantly increased several measures of manic and aggressive symptoms; these effects seemed to be independent of whether the participant had engaged in regular weight lifting or had used AASs illicitly in the past. These effects were largely confined to a small group of "responders"—but this group proved indistinguishable from "nonresponders" on demographic, psychological, physiological, and laboratory measures. The small number of responders in this analysis raises the possibility of a type II error—failure to reject the null hypothesis when a true difference exists. Also, the laboratory measures included total but not free testosterone, leaving open the small possibility that responders might have differed on the latter measure. However, such responses may be idiosyncratic and presently unpredictable.

Several limitations of the study should be considered. First, the crossover design raises the question of a carryover effect.^50,51 Although statistical methods have low power to detect a carryover effect, it seems unlikely: among participants who received testosterone in the first treatment period, all psychiatric measures returned to baseline at least 1 week before the beginning of the second treatment period (in other words, no significant differences were found between values at week 1 and at week 12 on any variable using paired t tests). Furthermore, in the analysis of the first treatment period only—which avoids the issue of a carryover effect—testosterone treatment differed significantly from placebo treatment on the YMRS, and showed trends in the same direction on the PSAP and on manic and liking scores on the daily diaries. Nevertheless, a carryover effect cannot be completely excluded in the primary analysis.

Second, study measures did not produce uniformly positive findings: the self-rating scales (AQ and Symptom Checklist-90-R) and diaries completed by the participants' significant others showed few differences, comparable to those found on the YMRS, PSAP, and daily diaries. However, neither the AQ nor the Symptom Checklist-90-R rates hypomania,^44,45 and the AQ may also be less sensitive than other measures of aggression.⁵⁷ The significant others' ratings also seemed to be insensitive, likely as a result of carelessness or noncompliance. Therefore, the YMRS, PSAP, and daily diaries were probably better adapted to measuring effects of testosterone.

Another possible criticism is that some participants might have guessed when they were receiving testosterone and hence displayed bias because of expectational effects.³¹ However, several observations weigh against this possibility. First, individuals previously experienced with AAS use or with regular weight lifting would presumably be more sensitive to expectational effects than those with no such experience. Yet, as mentioned previously, neither previous AAS use nor regular weight lifting was significantly associated with manic or aggressive responses to testosterone. Second, expectational bias would not easily explain the testosterone treatment effect on the PSAP, in which the participant was not informed that he was being tested for aggression. Third, there was no significant correlation between change in lean body mass and changes on the 8 psychiatric measures shown in Table 2 (P>.10 in all cases by Spearman rank correlation), suggesting that increased size and strength were not associated with elevated mood and aggression scores. These reasons notwithstanding, we cannot exclude the possibility of expectational bias.

Conversely, several methodological limitations may have caused us to underestimate the psychiatric effects of testosterone treatment. First, the modest dose and duration of testosterone treatment in our study, chosen for considerations of safety, did not match the high doses of AAS, frequently comprising several agents taken simultaneously and often ingested, by illicit users.^5-7 Second, we excluded prospective participants with a history of a major psychiatric disorder, but illicit users do not select themselves with similar care. Third, study participants were not permitted to use illicit drugs during the study, and none regularly consumed large amounts of alcohol. Illicit AAS users, lacking such restrictions, might be at higher risk for psychiatric effects than participants taking testosterone alone. Fourth, we withdrew 2 participants from testosterone treatment during the study because of adverse psychiatric effects; had these participants hypothetically continued through the full protocol, the effect of testosterone treatment might have been greater. For these reasons, the findings of the present investigation may represent a "lower bound" for the true rate of psychiatric effects exhibited by AAS users in the field.

In summary, our findings augment the evidence that administration of 300 mg per week of testosterone or the equivalent produces few psychiatric effects,^{6,23,27-34,36} whereas dosages of 500 per week or more produce occasional prominent manic or hypomanic reactions (Table 4). Such reactions are probably more frequent "in the field" than in the laboratory and may represent an underrecognized public health problem. To understand the biological or psychological contexts for these seemingly idiosyncratic reactions, future investigators should consider using larger sample sizes, more sophisticated batteries of neuroendocrine measures, and more detailed assessments of baseline psychological and neuropsychological indices.

Accepted for publication March 5, 1999.

This study was supported in part by grant RO1-DA06543 from the National Institute on Drug Abuse, Rockville, Md.

We are indebted to Alicja Skupny and Jack Mendelson, MD, for the performance and interpretation of the neuroendocrine measures; Don H. Catlin, MD, and Caroline K. Hatton, PhD, of the UCLA Olympic Analytical Laboratory, Los Angeles, Calif, for urinary anabolic-androgenic steroid testing; Pharmacia & Upjohn Co, Bridgewater, NJ, for supplying testosterone cypionate and matched placebo; David Katz, MD, for assistance with the design of the study and commentary on the manuscript; David Amato, PhD, and Garrett Fitzmaurice, PhD, for assistance with statistical analyses; and Paul Oliva and John Borowiecki for assistance with data collection and management.

Reprints: Harrison G. Pope, Jr, MD, MPH, Biological Psychiatry Laboratory, McLean Hospital, 115 Mill St, Belmont, MA 02178 (e-mail: pope@mclean.harvard.edu).