Doubts a Classic Lancet Study Showed Psychotherapy Improves Survival of Cancer Patients

bjhpA 2013 article in British Journal of Health Psychology by David Spiegel renews claims that his 1989 Lancet study demonstrated

“that women with metastatic breast cancer randomized to a year of weekly supportive-expressive group therapy lived 18 months longer than control patients.”

The Lancet article is

Spiegel D, Bloom JR, Kraemer HC, Gottheil E Effect of psychosocial treatment on survival of patients with metastatic breast cancer. Lancet. 1989;2(8668):888891.

The abstract is available here  and the actual article is here.

The persistence of this claim encourages to women with breast cancer to go to support or therapy groups with the expectation that doing so will increase their survival time. Many breast cancer patients already go to support groups believing they are improving their immune system, slowing progression of their disease and extending their survival time.

If the renewed claims by David Spiegel are valid, they should be widely disseminated. Women with breast cancer should be provided with ready access to support groups and supportive expressive therapy.

However, if not valid, women and their healthcare providers should be cautioned that this is unsubstantiated and unproven medicine. Women might want to nonetheless go to groups to get support, the opportunity to provide other women with support, and get validation for their experiences. However, they should not go with the illusion that they are affecting the progression and outcome of cancer.cancer support

In this blog post I question claims that  the Lancet study demonstrated that going to support groups or group therapy increased survival time.

The 1989 study was accompanied by an editorial stating

One of the central planks of the alternative medicine platform that mind can conquer matter and that the patient, adequately armed with the right attitudes of mind, can fight off the disease indefinitely.

The editorial at first suggested that readers adopt caution and skepticism in evaluating the claims of the article.

Results will undoubtedly reinforce the prejudice of all those who subscribe wholeheartedly to the mind over matter nexus-they do not even need to read the paper critically and their response will be “I told you so”. However, we would encourage our readers to adopt the same healthy scepticism about these claims as they would about the claims of any other breakthrough in the therapy of metastatic cancer.

However, the editorial continued

The methods adopted by the Californian workers seem to be beyond criticism, so they have indeed subjected their claims to the hazards of refutation. Careful reading of the entry criteria and prognostic categories within the randomised groups fails to demonstrate any important systematic or random bias that could affect outcome.

Why this study is not credible evidence that supportive expressive therapy increases survival time in metastatic breast cancer

The 1989 study met criteria for having a high risk of bias as evidence for effects of psychotherapy on survival time. The biases come from not having survival as its primary outcome at the outset, and not designating a time point for assessing ieffects on survival. If an effect on survival could not be observed at one time point, another could be examined, but this risks capitalizing on chance. That is why clinical trials are now required to pre-register time points for follow up, before the first patient is entered into a trial

There was no mechanism posited by which a survival effect should even have been anticipated.

The trial had a modest number of patients enrolled, 50 in the intervention group and 36 in the control group. It was thus grossly underpowered to detect an effect on survival. Claims of unanticipated strong effects being observed in an underpowered trial are usually false positives.

One of the original study’s authors, Helene Kraemer later published cogent arguments that clinical trials which such small numbers of patients should be excluded from meta-analyses because of their high likelihood of publication bias and repeated demonstration that results of such trials do not generalize to what is obtained with larger samples.

Lower risk of bias would come from a larger trial in which survival was identified as its primary outcome at the outset and a set time at which effects on survival would be assessed. Until results from such trials become available, no decision should be made about the clinical or scientific status of results obtained with a small trial, which should be treated as having a high likelihood of having occurred by chance or by methodological flaws.

Eventually, there were larger trials with survival as a declared outcome and a set point for assessing differences between intervention and control groups, but they did not find an effect on survival. These trials included one conducted by Spiegel and another on which he served as a consultant.

The editorial that accompanied the 1989 study put a lot of stock in the initial equivalency of the intervention and control group having been established by randomization.

However, the benefits of randomization cannot be assured with such small groups, where measured and unmeasured differences in key variables can determine results, either individually or in combination. The small group sizes are insufficient to test statistically for such influences.

A later study by Spiegel and colleagues demonstrated the powerful effects of baseline differences between groups on outcomes even in samples more than twice as large.

The 1989 article claims that the intervention group lived twice as long, but that is mean survival and is misleading. There are a number of measures of central tendency, by which the overall group outcomes can be characterized: mean, median, and mode.

Mean survival times for small groups are particularly vulnerable to outliers, which is why medians are generally considered more appropriate. Spiegel later revealed that the median survival in both groups was about 20 months. This should have been reported in the original article. Presumably, this difference i would be statistically nonsignificant.

survival curve-page-0

From Lancet 1989

The table of survival curves from the original article is reproduced here.

Few of the 1000+ people who cited this article seem to have noticed something strange about these survival curves of comparable deaths in the two groups. This should have raised red flags and begs an explanation.

Up until 20 months after randomization, the survival curves for Spiegel’s experimental and control groups were virtually superimposable. The control group was smaller at the outset (36 patients), and only 12 patients remained alive at 20 months when the survival curves diverge.

Note the inexplicable sudden drop in survival of the control group beyond 20 months. I would challenge anyone to find such a drop at this point in any clinical trial of a medical intervention for metastatic breast cancer.

Epidemiologist Bernie Fox and Onoologist Wallace Sampson were among the few observing  that the survival observed in the control group is significantly different from larger population of metastatic breast cancer patients from which it was drawn, where a 32% survival at 5–10 years is observed. Only 2.8% of Spiegel’s control group survived that time frame.

Any differences observed between the intervention and control group in survival could be due simply to the deviance of the control group, not the efficacy of the intervention.

In the subsequent attempts at replication, survival curves of intervention and control groups resemble that from Spiegel et al.’s intervention group, while none approximating the control group. This adds to the weight of evidence that the control group is deviant and the intervention inert.

The logic of interpreting tclinical trials requires the assumption that the outcomes for the control group would’ve been the same if it had received the intervention. This is obviously not a tenable assumption for this trial.

I doubt very much that were this a clinical trial of a medical intervention for metastatic breast cancer, that results would have been published in Lancet because of the deviancy if the control group, the small sample size, and the lack of specification of survival has a primary outcome.

Thus, Spiegel et al.’s 1989 findings were due to anomalous outcomes for the control group, not improvements in the intervention group for which no evidence exists.

These criticisms are not new and are contained in a number of published sources.* The article in British Journal of Health Psychology nowhere cites these criticisms.

Based on these criticisms, the American Cancer Society (ACS) revised its statement about the efficacy of support groups and psychotherapy:

The research is clear that support groups can affect quality of life, but the available scientific evidence does not support the idea that support groups or other forms of mental health therapy can by themselves help people with cancer live longer.

The UK National Health Service (NHS) recently commissioned a review and issued a statement about misleading claims in a journal article about an inexpensive blood test for risk of postpartum depression that were uncritically picked up by the media. The NHS should now issue a statement similar  to that of the ACS cautioning women of the lack of evidence that support groups increase survival or commission a review of its own of these claims and then issue a statement. Otherwise, the danger is that consumers will be misled by the abstract of the British Journal of Health Psychology article or by churnalled media reports.

For further resources, including abstracts, PDFs, and slide shows, go to here.


Coyne, J. C., Stefanek, M., & Palmer, S. C. (2007). Psychotherapy and survival in cancer: The conflict between hope and evidence. Psychological Bulletin, 133, 367-394. Available here.

Coyne, J. C., Thombs, B. D., Stefanek, M., & Palmer, S. C. (2009). Time to let go of the illusion that psychotherapy extends the survival of cancer patients: Reply to Kraemer, Kuchler, and Spiegel (2009). Psychological Bulletin, 135(2), 179-182.

Fox, B. H. (1998). A hypothesis about Spiegal et al.’s 1989 paper on psychosocial intervention and breast cancer survival. Psycho-Oncology, 7, 361–370.

Fox, B. H. (1999). Clarification regarding comments about a hypothesis. Psycho-Oncology, 8, 366-367.

Sampson, W. (2002). Controversies in cancer and the mind: Effects of psychosocial support. Seminars in Oncology, 29, 595-600.