Margot’s lover in La Reine Margot was one of the Huguenots who survived the massacre set in train by her brother Charles IX on St Bartholomew’s Day in Paris in 1572.
There are many politicians, bureaucrats, doctors and others, the Royalists, in a position to make a difference who know that psychotropic drugs can cause suicide or other serious problems but who instead attempt to close down any discussion of these issues. The Huguenots (patients) must not be alerted to what is going on. How it works is this – someone like me gets told by a Professor of Psychiatry you can’t say this to non-academic psychiatrists, or gets told by a psychiatrist you can’t say this to primary care doctors, or gets told by primary care doctors you can’t say this to patients. Its a control thing.
It can get very personal. Nearly 10 years ago the question of suicide on antidepressants came up in Ireland. I volunteered to debate with anyone on the topic. A number of senior figures declined. Finally two Irish academics accepted – with apparently extensive ad hominem backup from pharmaceutical companies for one of them in particular. They lost the debate. But to this day the Irish Psychiatric Association (IPA) goes out of its way to deny a link even though companies are legally obliged to state their drugs can cause suicide.
The driving force behind the creation of the IPA and its first President was a close friend over three decades. He is someone who knows that psychotropic drugs can cause suicide but faced with the Shane Clancy homicide-suicide was part of the public face of Irish psychiatry’s dismissal of a possible link to treatment – see Professional Suicide – The Clancy Case. Here was the perfect opportunity to open up a back channel conversation but all I got was a dead phone and since then nothing. Recently members of the IPA seem likely to have been involved in an extraordinary attempt to get the Clancy inquest re-opened. The request was turned down by the attorney-general.
It is against this background that the PLoS Medicine exchanges below around this Mortality in Schizophrenia article need to be interpreted. The review process in PLoS is typically open – when I review for this journal the authors know my name. But in this case I have no idea who the reviewers are or even which editor is commenting anonymously at the end. The first submission was turned down; let us pick up the story with the appeal. For first readers of this blog, the rules outlined in La Reine Margot for academic conversations apply. The other point to note is that there may be off the record comments to the editor that have not been forwarded to us – and these can be ad hominem in any kind of way – as may be reflected in the comments of the anonymous editor. If someone like Herb Meltzer (see La Reine Margot) was the reviewer (an obvious choice of reviewer) its easy to imagine what the comments might have been and there is no way for me to find out what might be happening.
We wish to appeal the decision in the case of this article. The basis for the appeal is that almost all the points raised by the reviewers, particularly the methodological ones, have been taken care of in the revised draft. The only other points are ones on which the reviewers have speculated rather than offered data or knowledge – viz there is no early intervention or outreach service here that might have contributed to the data we report, and there were no restrictions on the liberty of patients 100 years ago that might have led to the results then, and no reason to believe the illness was less severe then than now.
The methodological issues raised by the reviewers have greatly enhanced the paper and given that the findings are so striking and could contribute significantly to public health we still believe that publication in a high impact journal is called for.
Our responses to each point raised by each reviewer are given in bold below. We also include a tracked copy of all changes (Healy et al… Tracked) and a clean copy of the revised article (Healy et al … Clean).
Mortality in Schizophrenia and Related Psychoses: Two Cohorts, 1875-1924 & 1994-2010, Compared
The strengths of this paper are:
This paper can be improved by:
This is discussed briefly in the text – but is not a factor in our data. We have no early intervention service and no reason to believe first episodes are not being admitted here.
We have discussed this and in particular the validity of suicide diagnoses in both historical and contemporary cohorts in more detail
We have written an entire paper on this – it doesn’t seem appropriate to go into too much detail
We agree and have attempted to make this clearer. This problem is overcome in the survival analysis that has also been added to the current paper.
This paper addresses an important question for public health – whether there have been changes in standardized mortality ratios (SMRs) in patients with schizophrenia compared with the general population. This particular study examines case records of two cohorts from Wales, and concludes that there was been an increase in the SMR in the newer cohort, which can be mostly attributed to high SMRs for suicide.
There appear to be three main problems with the analyses and conclusions. First, as it is likely that more severe cases of schizophrenia were hospitalized in the newer cohort, the patients groups are not really comparable in this way. There is a 70 year gap between the last patients admitted in the older cohort and those admitted in the newer cohort – and over this time, the threshold criteria for hospitalization will have dramatically changed.
We have discussed this in detail and offered three reasons in the text as to why this is unlikely to be the case – the cases superficially were more severe historically, patients had relief from distress with medication today, and more patients were admitted voluntarily today.
Second, absolute numbers particularly in the newer cohort are very small, and to base the main conclusions of this paper on 13 deaths (for the diagnoses F20/25) in the newer cohort seems problematic. Hence, the CIs actually overlap for the main outcomes, so the authors are overstating their findings. So the average 5 year SMR for schizophrenia for both genders in the historical cohort was 4.0 (3.5-5.0) and 4.5 (3.7-5.6) in the newer one.
We have introduced more confidence intervals and modified out discussion in the abstract and discussion sections accordingly. The sample size is of lesser importance given that our findings are consistent with all other data in the field – the merit our study adds is its cohort nature makes it easier to see where risks arise
Finally, the authors attribute some of the high SMR to antipsychotic medication. This seems overly simplistic – many other possible reasons could explain the high SMR (to increasing stigma, lack of family support, untreated comorbid depression). Schizophrenia is a heterogeneous disease, and therefore the potential problems with diagnosis, changes in etiology, and treatment over the 70+ time periods between these two cohorts mean that in my view these data are at most hypothesis generating, but cannot be presented as definitive findings.
We have discussed this in detail in the revision. We offer reasons why medication is the most likely option – but have made this suggestion more tentative.
Two methodological queries – it is unclear from the text how SMRs were calculated for the newer cohort. Ideally, SMRs for each cause of death for every year by age band should be calculated. The text reads as if the authors took average SMRs for each cause of death over 1994-2010. I don’t think this is correct as there have been changes to suicide SMRs over this period, and the limited power of this study would argue against using averages in this way.
We have calculated SMRs as suggested standardized by age, sex and year for suicide. There are too few deaths from other causes to make the exercise reasonable
In addition, I was unclear about the extent of missing data. The authors state that 359 died in the community. How many of these did not have PM data on?
We have clarified this in the text.
This is very interesting and well written report. I have only few comments and suggestions.
page 5. Paragraph starting “All deaths in hospital (N=513) were recorded in…”. I suggest considering to move this this paragraph to “Results” section.
We have revamped this section completely
page 6. “When computing standardized mortality rates for years 2, 3, 4 and 5″, Does this mean “years after first admission”?
Yes. Again this is clarified
page 7. “There is a clear difference in age of death in those dying from suicide compared…” How was censoring treated? Instead of reporting mean age of death authors should apply survival analysis (Poisson regression or Cox’s PH model) and report RR or similar measures when comparing populations. ANOVA is not suitable for this kind of data.
We agree ANOVA is not suitable. As regards Table 3, we have given this issue a great deal of thought and have reorganized the material by age and diagnosis in Tables 3A and 3B
We have taken up the proposed survival analysis and this has added greatly to the paper. This has led to additions to the methods, results and discussion sections.
page 12. Paragraph starting “In the Tiihonen study, the best outcomes were found for patients on…”. This part of the discussion is not based on any results given in this MS, because authors used no medication data. This is more speculation than discussion based on facts.
Agree – have removed entirely
Tables 1, 2, and 3. These tables contain not much information. Authors should consider replacing them with results of survival analysis. E.g. Table 3 could be replaced with Age-Period type Poisson regression analyses. ( http://bendixcarstensen.com/APC/ )
As outlined we have replaced Table 3 and introduced Figure 1 – giving the survival analysis.
This is a well designed trial though there are potential avenues that could have improved the standard of the study. I am very worried by the censoring of individuals in the contemporary cohort. In particular the all column indicates 355 psychoses, with 33 causes of death however how many of these had 10 year follow-up data is not indicated by the authors.
We have clarified this issue and taken it into account A very large proportion had 5 year follow-up data.
1. In England and Wales there has been a full death registration system for some time. Individuals that left the asylum back to the community would have been registered as the deaths in the national system. The exact date and cause of death could have been obtained for all these individuals.
This is not reasonable. Because of the Welsh nature of the sample, we have 67 John Jones in our records. There is no way to get reasonable certainty as to time and cause of death in the missing sample. Our approach offers a safe over-estimate for likely mortality.
2. The historical cohort is obviously complete in terms of death, however the contemporary cohort has not yet reached the 5 year mark. Therefore the authors need to detail the size of the cohort by first admission year so that the reader can understand how many people are censored. It doesn’t affect the person years analysis, but the reader needs to know where the information is coming from. However the contemporary cohort is poorly explained with 1994-2010 in the drawing of the database, but 2008 for the mortality and then an indication of 10 year mortality which clearly has only been seen for a minority.
We have done this. Owing to the marked drop in incidence of schizophrenia from 2005, it will be seen in the data that fewer patients than might be expected are lost to follow up.
3. This leads to a more important comment about the analysis, not using person years has strong assumptions for the individuals for whom deaths were not observed. I’m afraid I couldn’t understand what the authors had actually done from their methods section. As far as I can tell they took just a count of the total number who were alive at the time of risk but excluded all those who died some unknown time in the future and then counted deaths that year. How is this count then adjusted for incomplete person years by the deaths?
We have now supplied a person years analysis. As regards adjustment for incomplete person years we have done a sensitivity analysis and refer to this in both results and discussion.
4. I am uncertain why the suicide deaths being uneven is a problem as there is not distributional assumption within a simple person-years analysis. Other analyses of rates do require distributions of deaths, but the current analysis doesn’t appear to.
We have done a person years analysis as suggested
5. The incidence data is not well explained as there is little information about the relative size of the general population that was used to calculate this information. The publication does go into this, but the aspect of severity gets lost between the two publications.
We have picked up the issue of severity in the discussion in response to the point made here and by reviewer #2
6. The severity of the mental health problems may well be very different in the two time periods therefore reflecting a different mortality risk.
We have discussed this further in detail in the discussion section – it seems most unlikely that patients in the modern period are more severely ill than the historical period
7. Was 2008 used for deaths in 2009-2010? When was you last death that was included?
The last death comes from 2009. We have managed the issue of loss to follow up by including a survival analysis.
8. The tables should have the condition not the ICD10 code as the heading.
They do now
9. The contemporary cohort is not well enough matured to analyse the data in this way. When the majority of your age groups have <5 deaths it is unlikely that any of the results are particularly well estimated. This is reflected within the confidence intervals.
ONS have rebanded the age-groupings and we have followed suit. It gives > 5 deaths in most age groups.
10. The authors have provided very few confidence intervals. This would clearly have shown such a large error rate.
We originally provided the confidence intervals for all key figures. We have now provided confidence intervals for all
11. The authors can actually calculate the years of life lost from their data which would aid their conclusions.
We have taken up this suggestion. This seems appropriate for the history cohort but less appropriate for the contemporary cohort for reasons outlined in the text.
12. The authors need to make sure that the conclusions take into account the size of the confidence interval in the discussion.
We have now done so
13. The risk of suicide is complicated by the issue that to have suicide as a cause of death you need to have opportunity to commit suicide. It may well be that registration of suicidal tendencies in the historic cohort meant that suicide opportunities were completely withdrawn using methods we would find unacceptable now and there is a benefit/harm balance to be had.
We have specifically dealt with this issue in the discussion. There was not an unacceptable deprivation of liberty in the historical period
14. The confidence interval for the SMR in the discussion appears to go to 0.0 this is highly unusual and seems at odds with the properties of the SMR. Is this correct?
We made a mistake here and this point has been taken care of
15. It is interesting that after removing suicides there is relative little excess risk of death. This needs more discussion.
We agree – this is the key point in the paper. We have discussed but don’t want to speculate too much
16. It is unlikely that short term mortality from cancer is related to a recent admission in either cohort as the majority of the cancers mentioned here have a much longer natural history.
We don’t relate cancer to recent admission. The discussion now makes this far more clear. But in addition we have used this issue to raise questions about estimates about loss of years of life
We got nowhere with these responses. The final put down came as an anonymous editors’ comments.
I have read through the comments and the manuscript, but echoing the previous comments I have serious concerns about the manuscript. The first issue, raised in the comments several times, is the hazard of making a diagnosis from chart review and comparing it with more recent criteria. There are many historical instances of patients being labeled as mentally ill when they did not fit in with societal norms (even a article in a psychiatry journal in the 1960s that described “angry young black man syndrome” during the beginning of the civil rights movement; young black men were hospitalized for months and placed on antipsychotics for this “psychotic condition”). The charts would have reflected the biases of the persons recording. It likely would be possible to make a diagnosis of schizophrenia in a classic case, but there are many more gray areas where it can be difficult to make a diagnosis even today with DSM criteria so the authors are likely not comparing apples with apples.
Second, the issue of looking at suicide rates within an institution vs in free-living patients is also highly problematic. Suicide requires means, and much of the point of the institutionalization then and now was to remove the means for a patient to commit suicide. The authors claim to have addressed this issue, saying “In addition, the monitoring of patients in the asylum was by the same methods used today. There was not an undue or unacceptable restriction of liberty by today’s standards.” I can’t imagine how they can say this categorically; before medications were available to constrain patients, physical restraints like straightjackets were used and these would hardly be acceptable by today’s standards. Regardless, a patient in an institution would have a much harder time getting access to the means to commit suicide than a person on the outside.
Third, the issue of selection bias because only the most severely ill patients are hospitalized now is a substantial one. I also am concerned about competing mortality risks given that TB was so frequent in the earlier population.
Finally, as raised in the reviewers’ comments, “the absolute numbers particularly in the newer cohort are very small, and to base the main conclusions of this paper on 13 deaths (for the diagnoses F20/25) in the newer cohort seems problematic. Hence, the CIs actually overlap for the main outcomes, so the authors are overstating their findings. So the average 5 year SMR for schizophrenia for both genders in the historical cohort was 4.0 (3.5-5.0) and 4.5 (3.7-5.6) in the newer one.” to which the authors replied: “We have introduced more confidence intervals and modified out discussion in the abstract and discussion sections accordingly. The sample size is of lesser importance given that our findings are consistent with all other data in the field – the merit our study adds is its cohort nature makes it easier to see.” The fact that their data are consistent (if they are) with previous studies doesn’t negate the fact that this study is based on extremely small numbers, and given all the above limitations the current data seem hard to support.
I emailed querying the anonymity of the editor and got a reply telling me that’s just the way it was.
From: “David Healy”
Date: 10 July 2012 19:43
Subject: Re: PMEDICINE-D-11-03195R2 Decision
To: PLoS Medicine
I am not raising these points to re-open the argument. I will take the paper elsewhere.
But I have already written a piece for History of Psychiatry on some of the issues raised by historical epidemiology of this type and the first point raised under Editor’s comments is one to add to the paper before it goes off. It would be better to attribute it to a person rather than to Editors PLoS Medicine and I wonder if you can help with this.
Here’s why its wrong. The paragraph refers to the incarceration of black men in the 1960s (there is a recent book on this topic which may have led to the mistake being made here).
The reason we know these black men were not schizophrenic is we have case notes extending over the entire time of their incarceration which retrospectively make it clear they did not have schizophrenia. Without the benefit of these records we would have no way of knowing.
But all the diagnoses made in N Wales in both historical and contemporary cases were made with this benefit – often with the benefit of 30-40 years worth of records.
Your reviewer/ editor has completely misunderstood the basis for diagnosis or is reading their own worries or prejudices into what is here rather than reading the paper.
Similarly with point 2, psychotic patients in North Wales a century ago spent the vast majority of their time on farms or in knitting rooms even when actively psychotic. The idea that they were straitjacketed is “lunatic”. We have the records to show this is wrong.
At least one of your reviewers/ editors appears to have a One Flew Over the Cuckoo’s Nest image drawn from US institutions in the 1950s as to the clinical population we are describing here.
Despite making it clear in my responses that today’s patients in North Wales are less severely ill than patients 100 years ago, the third comment reiterates a point – that may be true for the US today but is emphatically not true for North Wales.
I’m used to reviewers getting our data wrong. But 4 different journals have published our papers now and I’ve presented this at academic meetings involving epidemiologists and historians worldwide without significant criticism of the methods involved.
This is why its so disheartening to see PLoS get some of the key issues so badly wrong.
The issues here are largely about control. A lot of people in the system want to remain in control (they are Royalists). They do not want publication of data, they do not want the Bible put into the hands of the common people, who “would not understand”. Who would not see things the way we do.
By generating our own data RxISK.org will pose an interesting challenge to the Royalists.Share this: