Adjunctive Aspirin vs Placebo in Patients With Schizophrenia: Results of Two Randomized Controlled Trials

Abstract

Two previous randomized controlled trials (RCTs) suggested that adjunctive aspirin is efficacious in treating schizophrenia. We conducted two 16-week double-blind randomized placebo-controlled RCTs of adjunctive 1000 mg aspirin vs placebo in schizophrenia. Study 1 included 200 patients, with Positive and Negative Syndrome Scale (PANSS) total score as the primary outcome. Study 2 included 160 patients with C-reactive protein (CRP) >1 mg/L at baseline; the primary outcome was PANSS-positive score. Dropout rates for aspirin/placebo were 12% in study 1 and 20% in study 2. Differences in outcome between aspirin and placebo were calculated with linear regression, adjusting for the baseline value of the outcome. No statistically significant between-group differences were found in primary or secondary outcomes in either study. Study 1: mean difference in PANSS at 16 weeks was −3.9 (95% CI: −8.4 to 0.5, P = .10, effect size (ES) = −0.25) and at 8 weeks was −3.5 (95% CI: −7.5 to 0.5, P = .11, ES = −0.22). Study 2: mean difference in PANSS at 16 weeks was 0.3 (95% CI: −4.1 to 4.7, P = .90, ES = 0.02) and in positive PANSS was 0.5 (95% CI: −1.0 to 2.1, P = .50, ES = 0.11). A meta-analysis of these data with the existing studies, excluding one with large baseline differences in total PANSS, found that the overall estimate of the effect of adjunctive aspirin on the PANSS total score comparing group means at the end of the study was −2.9 (95% CI: −6.6 to 0.7; P = .21), favoring aspirin. Our studies and meta-analysis failed to find a statistically significant improvement in the symptoms of schizophrenia from adjunctive aspirin therapy in comparison to placebo in schizophrenia. Trial registration: study 1: Clinicaltrials.gov: NCT01320982; study 2 (high CRP): EudraCT Number: 2014-000757-36.

Introduction

The inflammatory hypothesis for schizophrenia has been supported by evidence from basic science, epidemiological associations, and biomarker studies.¹ Several drugs with anti-inflammatory properties have been tested as potential treatments for schizophrenia, with mixed results.^2,3 Aspirin (acetylsalicylic acid) irreversibly inhibits cyclooxygenase-1 (COX-1) and modifies the enzymatic activity of COX-2, thus inhibiting the formation of prostaglandins and reducing the inflammatory reaction. To our knowledge, there are only 2 published randomized controlled trials (RCTs) administering aspirin to patients with schizophrenia,^4,5 both showing improvement in total and positive Positive and Negative Syndrome Scale (PANSS) scores. A 2019 Cochrane meta-analysis of these 2 studies concluded that the evidence was “weak and inconclusive.” ⁶

We conducted 2 RCTs administering aspirin to patients with schizophrenia, one to all comers, the second to patients with significant positive symptoms and plasma C-reactive protein (CRP) >1 mg/L. We then conducted a meta-analysis including the results of these studies, together with the previous studies on the topic.

Methods

Study Population (Study 1)

Subjects, either inpatients (≥ 3 days after admission) or outpatients, were recruited from 18 sites in Romania and 1 site in the Republic of Moldova between January and June 2011. Participants were eligible if they were aged 18–65 years, met Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision criteria for schizophrenia or schizoaffective disorder and had at least 2 prior psychotic episodes and/or had been continually ill for at least 6 months. Other inclusion criteria were the use of an antipsychotic drug for at least 2 weeks prior to the baseline visit; a score of ≥4 (moderate or worse) on the PANSS items of delusions, hallucinatory behaviors, conceptual disorganization, and suspiciousness/persecution and/or a total PANSS negative symptoms score of ≥18 and a Clinical Global Impression Scale-Severity (CGI-S) score of ≥4 (moderate or worse).

Study participants were randomized to receive aspirin 1000 mg (n = 100) or placebo (n = 100) for 16 weeks. Because aspirin can cause gastric irritation, participants receiving aspirin also received pantoprazole 40 mg, and patients in the placebo group received a pantoprazole placebo. This study was part of a larger, 4-arm clinical trial, with patients randomized to aspirin (n = 100), minocycline (n = 100), pramipexole (n = 100), and placebo (n = 100). While this design has many advantages, it also has a higher probability of finding a false positive (rejection of the null hypothesis when the null hypothesis is true) compared to 3 independent studies each assigning subjects 1:1 to treatment and placebo groups. To control for the combined error rate from testing the 3 null hypotheses, we used a Sidak adjustment for all reports from this trial. Each of these compounds has a different mechanism; hence, each arm is being published separately, with its own introduction and discussion. The results of the minocycline arm have been published,⁷ and the results of the pramipexole arm are being prepared.

Sample Size and Statistical Power

Having 100 subjects in each of the 4 groups permitted power of >85% to detect a medium (d = 0.5) or larger effect size (ES) response, corresponding to a mean 15% improvement from baseline within the active treatment group compared to placebo.

Study Population (Study 2)

Subjects were either inpatients (≥3 days after admission) or outpatients, recruited from 30 sites in Romania from October 2014 to March 2016. Only subjects with plasma CRP >1 mg/L at baseline were included in the trial. Study participants were randomized to receive aspirin 1000 mg with pantoprazole 40 mg (n = 80) or aspirin placebo with pantoprazole placebo (n = 80) for a period of 16 weeks. Inclusion criteria were the same as study 1, except that, in study 2, the PANSS criteria included only patients with scores of ≥4 (moderate or worse) on 2 or more of the PANSS items of delusions, hallucinatory behaviors, conceptual disorganization, and suspiciousness/persecution without PANSS negative criteria.

Sample Size and Statistical Power

The sample size for study 2 was calculated based on early post hoc subgroup analyses of study 1, which suggested that aspirin improved PANSS-positive scores by an ES of 0.6 among participants in the highest tertile of baseline CRP. The interim analyses were not replicated in the full sample of study 1. When the statistician conducted a formal heterogeneity analysis for this paper with a test for effect modification by baseline CRP, the P-value for such test was .46, meaning that the null hypothesis of equal effects across CRP subgroups cannot be rejected. In order to obtain a result with 0.9 power, the sample size needed was 120 patients. Taking into account 25% dropout rates, we randomized 160 patients.

Study Medication Production, Randomization, and Masking

Medications for both studies were purchased from HEXAL and then overencapsulated and packaged by Sharp Clinical Services, formerly Bilcase GCS Ltd, in the United Kingdom. In both studies, a randomization list was created and provided by data management (Medistat; http://www.medistat.co.il/) to the study drug manufacturer, which assigned the medication based on the following blocks.

In study 1, randomization was done in blocks of 4, with 1:1:1:1 pramipexole/minocycline/acetylsalicylic acid/placebo. In Study 2, randomization was done in blocks of 2, with 1:1 aspirin/placebo. Raters were good clinical practice trained and board certified. All principal investigators had been involved in a number of previous international clinical trials. All raters who participated in the study had undergone PANSS training before the study. No formal PANSS training was done as part of the study, and interrater reliability was not assessed. In both studies, all study medications and placebo were packaged in identical appearing capsules.

Ethical Considerations

Both studies received approval from the national and local regulatory boards. Participants provided informed consent in accordance with the procedures outlined by the local institutional review board and were informed that they could withdraw from the study at any time. Both studies were performed following the Helsinki declaration.⁸

Study Assessments

For both studies, assessments were performed during clinical visits at baseline and 2, 5, 8, 12, and 16 weeks after randomization. The total PANSS score was the prespecified primary outcome for study 1 and the positive subscale score was the prespecified primary outcome for study 2, both at the end of the study. There was a mistake in the registration of study 1 in clinicaltrials.gov, where the primary outcome measure was erroneously recorded as both total PANSS at week 8 and total PANSS at week 16 (end of study). The primary outcome measure was only total PANSS at week 16. Additionally, there was a mistake in the registration of study 2 in the EU clinical trials registry, where the primary outcome measure was erroneously recorded as total PANSS. Other psychopathological symptoms (secondary outcomes) were assessed using the CGI-S and CGI-Improvement,⁹ and the Brief Assessment of Cognition in Schizophrenia (BACS).¹⁰ As all patients were receiving antipsychotics, structured assessments of potential side effects caused by antipsychotics were performed using the Simpson–Angus Scale¹¹ and the Udvalg for Kliniske Undersogelser Side Effect Rating.¹² We did not administer a specific scale focused on the side effects of aspirin. In between visits, patients received a weekly phone call inquiring about medication adherence, adverse events, and concomitant medications.

Statistical Analyses

Data analyses for both studies were identical, with one exception. Since study 1 had a 4-arm design (3 treatment groups vs 1 placebo group), in order to control for multiple comparisons, all study 1 P-values and CIs were adjusted with a Sidak correction¹³ for 3 comparisons. For primary and secondary outcomes, the main intention-to-treat analysis to determine group effect was an ANCOVA with last observations carried forward (LOCF). ES was based on Cohen’s d¹⁴ using the LOCF sample. As a sensitivity analysis for the missing data assumptions of the LOCF ANCOVA, we conducted mixed models for repeated measures using all available data with no imputations.¹⁵ Mixed models had visit, group, and group-by-visit interaction as fixed effects, an unstructured covariance matrix (for repeated measures on the same individuals), and specified restricted maximum likelihood. Also, as a sensitivity analysis, we repeated the main analysis, excluding subjects who took less than 75% of their assigned study medications based on pill counts. Between-group differences in adverse events and use of concomitant medications during the study were tested using Fisher’s exact test. Post hoc analyses exploring moderation by relevant baseline variables were performed. All tests applied were 2-tailed with an α of .05. For each study, the number needed to treat (NNT) was calculated based on the number of subjects that had ≥20% reduction in total PANSS (intention-to-treat). Data were analyzed using Stata version 16.

Meta-Analysis

We conducted 2 meta-analyses on the effect of adjunctive aspirin therapy on schizophrenia incorporating the 4 available RCTs: Laan et al⁴, Attari et al⁵, and the 2 trials presented here (summarized in supplementary table 1). Both meta-analyses are based on the total PANSS scores at the end of the study using LOCF estimates. One used the mean between-group differences at the end of the study and a sensitivity analysis used the mean between-group differences in the changes from baseline, excluding the Attari study, which had large differences in between-group baseline total PANSS scores differences.

Results

Study 1

Two hundred patients were randomized in the study (figure 1), all baseline demographic and clinical characteristics were well balanced between the groups; mean total PANSS at baseline was 96.3 (table 1). In the intent-to-treat analysis (table 2), the baseline-adjusted 16-week difference between the aspirin and placebo groups for total PANSS was −3.9 (95% CI: −8.4 to 0.5, P = .10, ES = −0.25) and for positive PANSS was −1.0 (95% CI: −2.5 to 0.5, P = .30, ES = −0.13). Similar trends were seen in the mixed model sensitivity analysis (table 2), secondary outcomes (figure 2A; supplementary figure 1A for BACS subscales), and at week 16 time point (supplementary table 3; per protocol sample—supplementary table 4), where the aspirin group had slightly larger reductions of symptoms, but the differences were not statistically significant. Total PANSS and positive PANSS at each visit are shown in supplementary table 2. The average compliance during the 16 weeks was 91% in the aspirin group and 87% in the placebo group. A sensitivity analysis excluding the participants with compliance less than 75% did not materially change results (supplementary table 3). There were no between-group differences in antipsychotic-related adverse events (supplementary table 5) or concomitant medications (supplementary table 6A). Fifty-two percent of participants in the aspirin group and 44% of those in the placebo group reported at least one adverse event during the study (P = .69).

Fig. 1.

CONSORT diagram.

Table 1.

Baseline demographic and clinical characteristics

	Study 1—no CRP inclusion criteria				Study 2—high baseline CRP
	(n = 200; 100/group)				(n = 160; 80/group)
	Aspirin		Placebo		Aspirin		Placebo
	Mean or n	SD or %	Mean or n	SD or %	Mean or n	SD or %	Mean or n	SD or %
Demographic characteristics
Age, mean (SD), y	42.2	(10.7)	43.5	(9.7)	42.7	(10.0)	40.4	(10.2)
Male, n (%)	49	(49.0%)	54	(54.0%)	40	(50.0%)	43	(53.8%)
Marital status, n (%)
Never married	62	(62.0%)	50	(50.0%)	44	(55.0%)	53	(66.2%)
Presently married	13	(13.0%)	18	(18.0%)	18	(22.5%)	14	(17.5%)
Divorced/separated	22	(22.0%)	26	(26.0%)	17	(21.2%)	12	(15.0%)
Widowed	3	(3.0%)	6	(6.0%)	1	(1.2%)	1	(1.2%)
Formal education, n (%)
Illiterate							1	(1.2%)
1–8 years	13	(13.0%)	11	(11.0%)	15	(18.8%)	14	(17.5%)
8–16 years	83	(83.0%)	83	(83.0%)	62	(77.5%)	62	(77.5%)
>16 years	4	(4.0%)	6	(6.0%)	3	(3.8%)	3	(3.8%)
Clinical characteristics
Inpatient, n (%)	23	(23.0%)	14	(14.0%)	13	(16.2%)	9	(11.2%)
Schizophrenia, n (%)	88	(88.0%)	90	(90.0%)	70	(87.5%)	70	(87.5%)
Schizoaffective disorder, n (%)	12	(12.0%)	11	(11.0%)	11	(13.8%)	12	(15.0%)
Number of hospitalizations, mean (SD)	18.0	(22.9)	15.6	(15.3)	9.3	(10.2)	10.4	(18.0)
Age at onset of psychiatric illness, mean (SD), y	25.8	(8.6)	26.3	(7.7)	32.7	(9.5)	31.4	(9.9)
Total PANSS, mean (SD)	96.0	(13.5)	96.5	(16.0)	100.7	(17.1)	100.7	(16.6)
Positive PANSS, mean (SD)	24.2	(4.2)	23.6	(4.7)	26.1	(5.2)	25.9	(5.1)
General PANSS, mean (SD)	47.1	(8.4)	47.1	(9.5)	50.3	(9.8)	50.7	(9.6)
Negative PANSS, mean (SD)	24.7	(5.3)	25.8	(6.2)	24.2	(5.1)	24.1	(4.7)
CGI severity, mean (SD)	4.8	(0.8)	4.8	(0.8)	5.0	(0.7)	5.0	(0.8)
BACS composite Z-score, mean (SD)	−3.6	(1.7)	−3.6	(2.1)	−3.9	(1.8)	−4.0	(1.4)

Note: CRP, C-reactive protein; PANSS, Positive and Negative Syndrome Scale; CGI, Clinical Global Impression; BACS, Brief Assessment of Cognition in Schizophrenia.

Table 2.

Primary and secondary endpoints at week 16

			Intention-to-treat analyses
	Mean (SD)		Effect size	Analysis of covariancea,b		MMRMb
	Aspirin	Placebo	Cohen's da	Difference (95% CI)	P-value	Difference (95% CI)	P-value
Study 1—no CRP inclusion criteriac	n = 91	n = 88	.	n = 200
Total PANSS	73.6 (14.8)	77.4 (18.9)	−0.25	−3.9 (−8.4, 0.5)	.10	−3.8 (−8.2, 0.6)	.11
Positive PANSS	17.1 (4.7)	17.6 (5.4)	−0.13	−1.0 (−2.5, 0.5)	.30	−1.0 (−2.5, 0.5)	.33
General PANSS	36.8 (7.6)	38.6 (10.2)	−0.23	−2.1 (−4.3, 0.2)	.08	. −2.1 (−4., 0.2)	.08
Negative PANSS	19.6 (4.6)	21.2 (5.8)	−0.27	−0.8 (−2.1, 0.6)	.44	−0.7 (−2.1, 0.6)	.47
CGI severity	3.7 (0.9)	3.9 (1.0)	−0.29	−0.3 (−0.5, 0.0)	.08	−0.2 (−0.5, 0.0)	.13
BACS composite Z-score	−2.7 (1.8)	−3.0 (2.1)	0.12	0.3 (−0.1, 0.6)	.13	0.3 (−0.1, 0.7)	.18
Study 2—high baseline CRP	n = 63	n = 64	.	n = 160
Total PANSS	77.5 (13.6)	77.9 (14.5)	0.02	0.3 (−4.1, 4.7)	.90	−0.6 (−5.1, 4.0)	.81
Positive PANSS	17.6 (4.2)	17.4 (4.3)	0.11	0.5 (−1.0, 2.1)	.50	0.1 (−1.3, 1.6)	.84
General PANSS	39.7 (7.5)	40.1 (7.8)	−0.06	−0.3 (−2.6, 1.9)	.77	−0.6 (−3.0, 1.8)	.61
Negative PANSS	20.3 (4.4)	20.4 (4.1)	0.04	0.1 (−1.0, 1.2)	.86	−0.1 (−1.3, 1.1)	.87
CGI severity	3.6 (1.0)	3.7 (0.9)	−0.08	−0.1 (−0.4, 0.2)	.67	−0.1 (−0.4, 0.2)	.52

Abbreviations are explained in the first footnote to table 1.

^aBased on the last observation carried forward.

^bAnalysis of covariance is the main analysis, and the mixed models for repeated measures (MMRM) is the sensitivity analysis. Differences are adjusted for the respective baseline value of each outcome. For PANSS and CGI, a negative difference indicates that the aspirin group had more improvement than the placebo group. For BACS, a positive difference indicates that the aspirin group had more improvement than the placebo group.

^cFor Study 1, all P-values are Sidak-corrected to account for the 4-arm design (3 between-group comparisons).

Fig. 2.

Outcomes over time according to randomization group.

Median (interquartile range [IQR]) plasma CRP levels at baseline were 2.5 (1.1, 5.2) in the aspirin group and 2.1 (0.7, 4.3) in the placebo group (supplementary table 7A). There were no between-group differences in CRP levels at the end of the study (P = .20 for Wilcoxon rank-sum test). In exploratory analyses, baseline CRP was not a moderator of the effect of aspirin (supplementary figure 2; likelihood ratio test P = .46). Further exploratory heterogeneity analyses also did not suggest moderation of the effect of aspirin by baseline PANSS, age, illness duration, gender, or polypharmacy (supplementary figure 3A).

Study 2

One hundred and sixty patients were randomized in the study (figure 1), baseline demographic and clinical characteristics were well balanced between the randomization groups. Mean PANSS total at baseline was 100.7 (table 1). In the intent-to-treat analysis (table 2), the baseline-adjusted 16-week difference between the aspirin and placebo groups in positive PANSS was 0.5 (95% CI: −1.0 to 2.1, P = .50, ES = 0.11) and 0.3 (95% CI: −4.1 to 4.7, P = .90, ES = 0.02) in PANSS total score. This was consistent with estimates from mixed models (table 2). There were no statistically significant differences in PANSS subscales between aspirin and placebo at earlier time points (supplementary table 2). Similarly, there were no statistically significant between-group differences in other secondary outcomes at the end of the study, ES ranging from −0.08 to 0.11, and no apparent differences between placebo and aspirin groups at any time during the study (figure 2B and supplementary figure 1B for BACS). The average compliance during the 16 weeks was 79% in the aspirin group and 80% in the placebo group. A sensitivity analysis excluding participants with compliance less than 75% did not materially change results (supplementary table 3). There were no between-group differences in adverse events (supplementary table 5) or concomitant medications (supplementary table 6B). Thirty percent of participants in the aspirin group and 28% of those in the placebo group reported any adverse event during the study (P = .86).

Median (IQR) CRP levels at baseline were 6.2 mg/L (4.0, 11.4) in the aspirin group and 6.8 mg/L (3.9, 14.3) in the placebo group (supplementary table 7B). There were no between-group differences in CRP levels at the end of the study (P = .37 for Wilcoxon rank-sum test). In exploratory analyses, baseline CRP was not a moderator of the effect of aspirin (supplementary figure 2; likelihood ratio test P = 0.57). Further exploratory heterogeneity analyses also did not suggest moderation of the effect of aspirin by baseline PANSS, age, illness duration, gender, or polypharmacy (supplementary figure 3B).

In study 1, the NNT was 10 (95% CI: −26 to 4) favoring aspirin and, in study 2, the NNT was 80 (95% CI: −6 to 7) favoring placebo. The dropout rates for aspirin/placebo were 9/12% in study 1 and 21/20% in study 2. In study 1, there were 2 severe adverse events (SAE) in the aspirin group and 1 SAE in the placebo group. In study 2, there were 2 SAEs in the aspirin group and 4 SAEs in the placebo group. There was no evidence of changes in the side effects caused by antipsychotic treatment (supplementary table 8).

Meta-Analysis

The combined sample size from the 4 studies in the meta-analyses was 490 subjects. The overall estimate of the effect of adjunctive aspirin on the PANSS total score comparing group means at the end of the study was −2.3 (95% CI: −6.3 to 1.6; P = 0.24), favoring aspirin (figure 3A). There was no substantial heterogeneity between studies, with an I² of 39% (Q(3) = 4.9, P = 0.18). The meta-analysis using the between-group differences in the changes from baseline excluded the Attari study due to very large differences in total PANSS total at baseline and found an overall estimate of the effect of aspirin on PANSS total to be −4.3 (95% CI −9.3 to 0.8; P = .10), also not statistically significant (Figure 3B). There was substantial heterogeneity between the studies, with an I² of 67% (Q(3) = 9.2, P = 0.03).

Fig. 3.

Meta-analysis for the effect of aspirin on Positive and Negative Syndrome Scale total.

Discussion

Aspirin reduces inflammation through the inactivation of the COX enzyme, which is needed for the synthesis of prostaglandins and thromboxanes,¹⁶ uncoupling oxidative phosphorylation in mitochondria, and modulation of signaling through NF-κB, which have all been implicated in schizophrenia.^17–19 This manuscript presents 2 clinical trials of adjunctive aspirin therapy, the first enrolling all qualifying and consenting patients, and the second enriched by including patients with elevated CRP levels, hence with potentially higher inflammatory activity. These data and the results of the meta-analysis are most consistent with a small, not statistically significant effect of aspirin on PANSS total score with the current sample size that is not moderated by CRP levels. This small effect is unlikely to be clinically significant unless biomarkers identify a subgroup of individuals who exhibit larger responses. The NNT was 10 in study 1 (95% CI: −26 to 4) favoring aspirin and 80 in study 2 (95% CI: −6 to 7) favoring placebo, both not significant.

Study 1 was designed to replicate findings from the trial by Laan et al, which, at the time of the conception of the trial, was the only published RCT testing aspirin for schizophrenia.⁴ Laan et al reported statistically significant benefits of aspirin vs placebo on total PANSS (ES = −0.47) and PANSS-positive scores (ES = −0.39). While directionally consistent with the Laan trial, none of our prespecified outcomes showed statistically significant differences in the aspirin vs placebo groups.

Study 2 was designed based on the hypothesis that participants with more inflammatory activity, as measured by baseline CRP >1 mg/L, would show a greater benefit from aspirin. However, that was not the case in our study 2, as the ES for the primary outcome of positive PANSS was 0.11, and none of the primary or secondary outcomes showed statistically significant differences in the aspirin vs placebo groups. Post hoc analyses exploring potential moderation by baseline CRP did not support the hypothesis that aspirin could be beneficial in subgroups with higher CRP. There was no evidence that aspirin reduced CRP levels compared to placebo, a finding similar to another short-term study administering aspirin to patients with bipolar disorder.²⁰

Our results are similar to a recently published study on the combination of salsalate 4g/d, omega-3 fatty acids, and a statin added on to antipsychotic treatment in a 12-week study, which showed no effect on persistent positive symptoms, negative symptoms, or cognitive impairments.²¹

Meta-Analyses

In order to provide context on the interpretation of the Attari trial, which had large between-group baseline differences in PANSS, we present 3 meta-analyses of the 4 published aspirin for schizophrenia RCTs. The change-from-baseline meta-analysis yields a PANSS mean difference of −4.3 (95% CI: −9.3 to 0.8) favoring aspirin with high heterogeneity (I² = 67%) due to the Attari trial. The PANSS total score was 106 in the combined aspirin groups compared to 98 in the control group at the end of the 6-week intervention (supplementary table 1). The most reliable estimate of the effect adjunctive aspirin on schizophrenia comes from our meta-analysis of end-of-study estimates (figure 3A). In this analysis, the 4 RCTs are in relative agreement (I² = 39%), with a mean difference in PANSS of −2.3 (95% CI: −6.3 to 1.6). That is equivalent to a Cohen’s d ES of −0.12 (95% CI: −0.38 to 0.14). When Attari is omitted, the results fall between the other methods (supplementary table 2). In other words, we could expect the true effect of aspirin to vary from slightly worsening symptoms (ES of 0.14) to a meaningful improvement of symptoms (ES of −0.38), and, at this time, we cannot rule out either with 95% certainty. Our meta-analysis showed the effect of aspirin to be 2.3–4.3 PANSS total points better than placebo. While neither was statistically significant, with larger sample sizes, this magnitude of effect would certainly be statistically significant. In perspective, it is important to note that, in studies including 135 000 patients,²² aspirin decreased the risk of stroke and myocardial infarction in high-risk patients. To have 80% power to detect a 5% reduction in PANSS compared to placebo (eg, −4 PANSS difference in aspirin vs placebo), a total sample size of about 524 subjects would be needed.

The results of these trials should be put into perspective. When the inflammatory theory was new, the initially published studies were on smaller samples and showed mainly positive findings. We are now in a later phase of research on this topic, where larger RCTs are performed and negative results or less positive results become published. This phenomenon, where positive results in early-phase, smaller studies are not later replicated in larger studies, or clearly replicated but with smaller ES, is common in many fields of medicine.²³

Side Effects

While we did not observe significant side effects of aspirin, our 2 short interventions were not adequately designed to capture this. A systematic review on the studies of aspirin for secondary prevention of ischemic stroke²⁴ on tens of thousands of patients using doses 1/20th to 1/3rd of what we used in our studies (1000 mg/day) reported a number needed to harm for aspirin of 62 (95% CI: 48.2–84.4).

Potential Confounders and Limitations

One potential concern in our study is that all of the patients who were getting aspirin also received a proton-pump inhibitor (pantoprazole), which has few interaction with other drugs,^25,26 so it is unlikely that it might have modified the level of antipsychotics. However, participants in our studies were chronic patients who have been ill for many years (study 1: mean of 16.9 years; study 2: mean of 9.5 years) with relatively high levels of symptoms.

The Laan trial treated patients early in the course of the disease (mean 3.7 years after the onset of illness) and found aspirin to have a beneficial effect. Although our study cannot rule out a beneficial effect of aspirin in prodromal or first-episode patients, we conducted a post hoc analysis of the effect of aspirin on total PANSS, stratifying by median age at baseline, and found no difference in treatment effect in either study (all treatment by age interactions P-values >.34). The more recently published Attari study on adjunctive aspirin also reported improvements in symptoms in younger participants, mean age 33⁵, although interpretation is clouded by the large baseline differences in baseline PANSS. A recent meta-analysis of anti-inflammatory drugs also concluded that anti-inflammatory agents might be more beneficial in first-episode psychosis or early-phase schizophrenia,²⁷ and it is conceivable that aspirin might benefit a pathological mechanism operative in the early stage of the disease. This suggests the need for future studies on younger patients early in the course of their illness. Celecoxib has also been tested for schizophrenia; some,^28–30 but not other studies^31,32 showed efficacy, a meta-analysis did not show significant improvement. Taken with the evidence suggesting a role for inflammation in the pathophysiology of schizophrenia, our results suggest the study of other anti-inflammatory agents or other approaches to target nonaspirin-sensitive constructs.

Overall, our results are consistent with a small not statistically significant effect of aspirin on PANSS total score that is not moderated by CRP levels. This small effect is unlikely to be clinically relevant unless biomarkers are developed that identify a subgroup of individuals who exhibit larger responses. These results, along with the risk of side effects, such as gastric hemorrhage, might be considered by physicians who treat patients with symptomatic chronic schizophrenia treated with antipsychotics, who have few treatment alternatives.

While there is a sound theoretical basis for an anti-inflammatory hypothesis in schizophrenia, we conclude that our studies and meta-analysis failed to find a statistically significant improvement of adjunctive aspirin therapy in comparison to placebo in schizophrenia.

Funding

This work was supported by the Stanley Medical Research Institute.