Landmark Psychiatry Papers
What counts as evidence that a treatment helps?
A treatment claim needs a comparison
Psychiatry advances when a persuasive clinical observation becomes a question that other investigators can test. Improvement is the beginning of that question. To interpret it, we need to know what would have happened under a different treatment, how improvement was measured, and whose experience the study represents.
This inaugural website issue considers three contributions to that task: Schou and colleagues’ controlled lithium investigation, DeRubeis and colleagues’ comparison of cognitive therapy with medication, and the first-phase CATIE antipsychotic trial. They are landmarks for different reasons. One strengthened an early therapeutic observation; another tested a disputed boundary between treatment modalities; the third made continuing treatment, with its benefits and burdens, the central outcome.
The purpose of revisiting them is critical appraisal. Their historical importance does not make their designs interchangeable or their conclusions universal.
1. Schou and colleagues, 1954: Testing improvement against the course of mania
Reference: Schou M, Juel-Nielsen N, Strömgren E, Voldby H. The treatment of manic psychoses by the administration of lithium salts. J Neurol Neurosurg Psychiatry. 1954;17(4):250–260. Original article. DOI: 10.1136/jnnp.17.4.250.
Design and findings. The Danish investigation included 38 patients with mania. It combined open treatment with blinded lithium–placebo periods, usually lasting two weeks, using random treatment switches. Daily observations recorded mood and motor activity on a three-point scale. This was not a uniform parallel-group randomized trial.
The investigators classified 14 patients as having a convincing lithium effect, 18 a possible effect, and six no beneficial effect. These were judgments incorporating illness course, not a modern response rate with a randomized control denominator. Toxicity was a substantial concern.
Contribution and limits. Its importance lies in the attempt to separate therapeutic change from spontaneous fluctuation and observer expectation. Small numbers, coarse ratings, mixed treatment arrangements, and potential carryover between periods limit causal precision. The investigation supported further testing of lithium in mania; it did not establish a modern maintenance regimen. Full text, especially pp. 250–256.
2. DeRubeis and colleagues, 2005: Comparing treatments without assuming equivalence
Reference: DeRubeis RJ, Hollon SD, Amsterdam JD, et al. Cognitive therapy vs medications in the treatment of moderate to severe depression. Arch Gen Psychiatry. 2005;62(4):409–416. DOI: 10.1001/archpsyc.62.4.409. Author-hosted full text.
Design and findings. At two research clinics, 240 adults were randomized to medication (120), cognitive therapy (60), or pill placebo (60). Active treatment lasted 16 weeks; placebo lasted eight. Pharmacotherapy began with paroxetine and permitted later augmentation. Outcome interviewers were masked.
At 16 weeks, 58% in each active arm met the study’s response criteria: medication 95% CI 48%–66%; cognitive therapy 45%–70%. Remission was 46% and 40%, respectively. Medication performed better on continuous symptom change at Vanderbilt; that advantage was not observed at Pennsylvania.
Contribution and limits. The trial challenged a simple severity-based exclusion of cognitive therapy. It did not prove equivalence. Results differed by site, and therapist expertise was a plausible explanation, not a randomized finding. There was no placebo comparison at 16 weeks. These selected outpatients cannot represent every clinical presentation. Methods and results, pp. 410–414.
3. Lieberman and colleagues, 2005: Asking whether treatment remains workable
Reference: Lieberman JA, Stroup TS, McEvoy JP, et al.; CATIE Investigators. Effectiveness of antipsychotic drugs in patients with chronic schizophrenia. N Engl J Med. 2005;353(12):1209–1223. DOI: 10.1056/NEJMoa051688.
Design. CATIE recruited 1,493 participants at 57 US sites. Its blinded first phase compared perphenazine with several second-generation antipsychotics for up to 18 months. The principal outcome was time to discontinuing the assigned drug for any reason. The original protocol paper explains the deliberate emphasis on effectiveness across varied settings.
Findings and limits. Among 1,432 participants who took medication, 1,061 (74%) discontinued their assigned drug before 18 months. Olanzapine had longer treatment continuation than quetiapine or risperidone, alongside greater weight and metabolic burdens. This was a trade-off, not an unqualified winner.
Stopping an assigned drug combines several reasons and does not necessarily mean stopping all treatment. Perphenazine allocation excluded participants with tardive dyskinesia; the comparison cannot establish that all older and newer drugs are equivalent. First-episode and treatment-resistant populations also require separate evidence. Original report.
A 2010 correction amended hospitalization data and the prolactin unit. Those hospitalization estimates are not reproduced here.
Three different questions, three different answers
| Study | Central evidentiary question | Inference to avoid |
|---|---|---|
| Schou, 1954 | Can observed change be distinguished from fluctuation and expectation? | Treating clinical classifications as a modern response-rate comparison |
| DeRubeis, 2005 | How do two active approaches compare under a specified protocol? | Reading a nonsignificant difference as proof of equivalence |
| CATIE, 2005 | Which assigned treatment remains acceptable and useful over time? | Reading discontinuation as a pure measure of symptom efficacy |
The following questions are an editorial framework for reading treatment research, rather than additional findings from these trials.
What exactly was compared?
An intervention is more than its name. A medication strategy includes dose adjustment, follow-up, and decisions about adverse effects. A psychotherapy includes its procedures, session frequency, and the people delivering it. Before asking whether two treatments differ, specify the versions being compared. A result about one treatment package cannot settle every comparison carrying the same labels.
Does the outcome match the claim?
Symptom change, remission, continuing treatment, and functioning are different outcomes. A study can be informative about one while leaving another unresolved. In appraisal, write the outcome into the conclusion: “improved this measure over this interval” is more precise than “worked.” Then ask whether that outcome captures what matters to the person receiving care.
How much uncertainty remains?
A point estimate is not the whole result. A confidence interval expresses uncertainty around an estimate under the analysis assumptions. Similar observed percentages do not show that meaningful differences have been excluded. Formal equivalence or noninferiority requires a design and prespecified margin suited to that question; failure to detect superiority does not supply them retrospectively.
Can a proposed explanation be separated from a finding?
Differences across sites can generate useful hypotheses about expertise, patient characteristics, or service organization. They do not automatically identify the cause of those differences. The same discipline applies to mechanisms: observing benefit does not, by itself, explain why a disorder occurs or how the intervention produces change.
What these landmarks contribute to clinical reasoning
The common contribution is a habit of making treatment claims conditional and inspectable. Specify the population, comparator, outcome, time horizon, and uncertainty before translating a result into a clinical discussion.
For a journal club, try rewriting one conclusion from each paper in a single sentence containing those five elements. Then identify the additional evidence needed to address durability, harms, everyday functioning, and applicability to a different patient group. That exercise preserves the achievement of the original investigation while making the remaining questions visible.
**Discussion question: When a treatment improves symptoms but is difficult to continue, which outcomes would help a patient and clinician judge its value together?**