‘Dangerous data’: drinking after dependence part 5. Sobells in the firing line
The temperature of the controversy peaks during a bitter and decades-long controversy over the Sobells’ findings. ‘Fraud’ claimed the detractors, but investigators disagreed. Mike Ashton of Drug and Alcohol Findings continues his serialisation of the project’s essay on the most controversial issue in alcohol treatment: whether dependent drinkers should always aim for abstinence.
Last episode broke off the story in the late 1970s when second– and third-year follow-ups of the Sobells’ US sample had shown that superior outcomes among patients considered suitable for and trained in controlled drinking were persistent and broadly based in their lives beyond drinking.
Just as with Davies’ research at the Maudsley in London (part 2), the most robust challenge to the Sobells’ findings came on the back of new data from the patients themselves in the form of another follow-up by a sceptical research team. Published in Science magazine in 1982, it cast doubt on the validity of the earlier findings – so much so that via the journalist Philip Boffey, co-author Irving Maltzman assured readers of the New York Times that “Beyond any reasonable doubt, it’s fraud.” Another co-author also told reporters their findings cast “grave doubt on the scientific integrity of the original research”. On the spectrum of academic criticism, the gravest of accusations was being levelled at the Sobells, one with its origins in serendipitous links between key players; see panel below.
During 1976 to 1979 – five to nine years after the patients had been treated at Patton hospital – Maltzman together with Jolyon West and lead author Mary Pendery managed to contact and (then or subsequently) interview all but one of the 19 survivors from among the key 20 in the Sobell study – the ones judged appropriate for and trained in controlled drinking. One task for the interview was to identify records which would confirm how the patients had fared, reducing reliance on memory and honesty. Though information was given on the patients up to “the end of 1981”, the report focused on contrasting the picture it pieced together of their drinking in the first and third years after discharge with the picture given for the same periods by the Sobells and by Glen Caddy.
Over those three years the critics identified just one of the 20 patients as having “succeeded at controlled drinking”. What was meant by this was unclear, and the report eludes direct comparison with the articles it purported to contradict because it is primarily a narrative of adverse events for each individual (such as heavy drinking, hospitalisations and illness), without these being organised as the Sobells and Caddy did into proportions of days abstinent, moderately drinking, drunk, or institutionalised due to drinking. That adverse events happened was not at issue, especially in the six months after discharge, when even according to the Sobells on a third of days patients allocated to the controlled-drinking programme drank heavily or were in hospital or jail due to drinking. In comparison to the Sobells’ methodology, highlighting the ‘bad days’ and inviting readers to judge the patients’ progress on these was bound to present the more unpalatable (but also more unrepresentative) picture the critics were looking for.2
Bullets lack a benchmark
The fundamental weakness of Pendery’s refutation study was that while it did partially document the progress of the key patients, it did not benchmark this against the other groups. In their reply, Mark and Linda Sobell actually admitted to many more hospitalisations for these patients than their critics had uncovered – but the ‘bad days’ spent in heavy drinking or institutionalised due to drinking were far fewer than among comparable patients allocated to conventional abstinence-oriented treatment. All the adverse events Pendery and colleagues reported could be accepted without affecting the Sobells’ core contention: that though the controlled-drinking patients (all with a poor prognosis) were not uniformly successful in overcoming their dependence, they did far better than their conventionally treated comparators.
Reasons given3 by Mary Pendery and colleagues for not reporting on control patients (even though they had also been followed up) would have needed to have been taken into account in adjusting and interpreting a comparison with the controlled-drinking patients, but seem well short of a justification for not making the comparison in the first place. Perhaps for the critics the clinching consideration was that “we are addressing the question of whether controlled drinking is itself a desirable treatment goal, not the question of whether the patients directed toward that goal fared better or worse than a control group that all agree fared badly” – reasoning which neglected the reality that no treatment or treatment goal is desirable in isolation, but only relative to the alternatives open to the patients. This same fundamental point was made by the most thorough of the investigations into the Sobells’ work. The Dickens inquiry (of which more below) argued that “drawing inferences from [these data] with respect to treatment effectiveness demands a comparison. Science, the activity, would have demanded such a comparison even though Science, the magazine, did not.”
That comment was made in respect of the most important of the available outcomes – patient deaths; it turned out to be one where lack of a comparison was clearly critical. Among the evidence cited by Pendery and colleagues were four “alcohol-related deaths” up to the end of their follow-up period in 1981 among patients trained in controlled drinking. But further investigation by the Sobells showed that the death record was actually worse among patients assigned to traditional abstinence-based treatment. The four allocated to training in controlled drinking had died on average about nine years after participating in the Sobells’ study, three for clearly alcohol-related reasons. But six comparison patients had died, on average about seven-and-a-half years post-participation, four clearly due to their drinking. If anything, training in controlled drinking had prevented early alcohol-related deaths.
The comparison did not stop CBS’s 60 Minutes TV programme broadcasting in March 1983 a tour of the graves of the four men, who according to the SAGE Encyclopedia of Alcohol were represented as having been killed by an irresponsible experiment: “60 Minutes failed to mention the six men who had died after undergoing the standard abstinence-based treatment or the poor outcomes that abstinence-based treatment programs have in general.”
One volley hits home
Where the Sobells had to bow to criticism was in respect of their assertion that “follow-up interviews were regularly conducted every 3–4 weeks” and that (free source at time of writing) “Subjects … were contacted for follow-up every 3–4 weeks for a period of 2 yr”. In fact, the schedule was not maintained. It was not a trivial point: at each contact a detailed drinking record since their last contact was sought from each patient, enabling the calculation of the proportion of days of controlled drinking, abstinence, or heavy drinking on which the outcomes of the study largely rested; the shorter the recall period, presumably the more accurate the data.
One of the investigations into the study found (1 2) that over the two-year follow-up contacts averaged about 15.4 It was far more than the “four or fewer times” Mary Pendery said “most” patients had been contacted, but well below the scheduled 24.5 The Sobells must have known they had not kept to schedule, said the critics, yet did not disclose this in their publications – an omission which formed the prime basis6 for alleging fraud in the form of fabrication of missing data, one majored on in an article, book and news report.
The mystery remains why, if the intention was fraud, the Sobells revealed in their second-year follow-up article (free source at time of writing) that “Five of the 69 subjects found were extremely difficult to locate for follow-up. Final data for these subjects was completed long after their designated follow-up intervals had expired” – a public admission that the follow-up schedule was not always achieved.
No fraud, say investigators
Investigations of the follow-up frequency issue uncovered (free source at the time of writing) error and carelessness, not fraud, and none said shortcomings invalidated the findings: “There was no convincing evidence that your admittedly erroneous estimates of the frequency of follow-up contacts significantly affected the conclusions drawn from this research.” It might be added that 15 follow-up contacts over two years would still make the study’s tracking of post-treatment drinking one of the most detailed ever conducted.
The judgement quoted in the previous paragraph came from the ethics committee of the American Psychological Association, which mounted one of the four inquiries into the Sobells’ scientific integrity. Similar territory was adjudicated on in a court case initiated by former study participants. The inquiries’ methods and findings have been endorsed by the Sobells (1, free source at the time of writing; 2), attacked by critics (1 2), and usefully reviewed by the independent voice of Ron Roizen.
First and most thorough it seems7 was an investigation by the Sobells’ then employers, the Canadian Addiction Research Foundation. Known after its chair as the Dickens Committee inquiry, it was conducted by four eminent academics, one of whom was at the time Canada’s minister of state for science and technology. Their report was delivered in 1982.
The following year the Sobells received the verdict of an investigator sent by the science committee of the US Congress to check for evidence of fraud. Hard on its heels, in 1984 came the report of an investigation overseen by a five-member panel consisting of senior staff from the US government’s drug and alcohol, health, and health research departments, known after the panel’s head as the Trachtenberg report. The final inquiry was mounted by the ethics committee of the American Psychological Association, which reported in 1984. Its instigation has variously been described as a complaint by the Sobells that their critics’ (Drs Maltzman and Pendery) accusations had damaged their reputations, a similar complaint in which the Sobells had been joined by the lead author of the third-year follow-up report, and a complaint in the opposite direction (free source at the time of writing) from Dr Irving Maltzman about the Sobells’ work.8
Last to come to a conclusion, and one of several legal proceedings, was a lawsuit against the State of California mounted by some of the participants in the Sobells’ study along with their relatives, alleging that the study had led to arrests, pain, public humiliation and four deaths, and that data had been “negligently or intentionally misrepresented and falsified”. Launched in 1983, it was dismissed (free source at the time of writing) by a judge in 1987.
Not a verdict on scientific validity
All the investigations focused on allegations of scientific fraud, most seriously in the form of intentional fabrication of evidence and a cover-up of this fabrication. None came near endorsing the charges. Errors, ambiguities, incomplete descriptions of methodology – these there were, but such shortcomings are common in substance use evaluation research, and not usually seen as indicative of fraud. As the Sobells saw it, in this case the heat had been turned up from routine and probably inconsequential shortcomings to far from routine fraud, because the notion that “chronic dependent alcoholics could successfully control their drinking was a terribly threatening idea at the time to alcoholism treatment personnel,” a threat which led critics spearheaded by Drs Maltzman and Pendery “to unleash this attack on this study as a way of disabusing anyone from believing that such ‘controlled drinking’ is possible”.
However, to conclude there was no fraud is not the same as declaring that methodologically, all was well. Regardless of the motivations of the accusers and the innocence of the accused, the possibility would remain that inadequacies in the research rendered the findings unreliable, and therefore also the interpretations and conclusions based on these findings. None of the investigations were geared to examining the methodology of the study as such, only in so far as it might pertain to allegations of intentional wrongdoing. Ironically, the follow-up on which the main critics rested much of their case helped by largely validating the Sobells’ records and published findings.
When closely compared, findings presented by the critics as “in marked contrast to the favorable controlled drinking outcomes reported by the Sobells and Caddy et al” were found9 to offer more confirmation than refutation, lending the weight of an independent follow-up to the original study. Another close comparison was conducted by the Dickens committee, which found that on the important issue of re-hospitalisation due to alcohol-related causes, “for each person … the Pendery et al. and the Sobells’ data coincided”.10 Rather than the data itself diverging wildly, the divergence was largely in how it was interpreted, itself dependent on what the interpreter wished to portray, and whether the contrast was made with the poorer performance of the era’s conventional treatment both in general and as applied to the controls in the Sobells’ study.
In his essay on the controversy, Ron Roizen explained how the same data can signify the opposite to observers on different sides of the abstinence-only divide. Evidence that many potential patients reject abstinence may be seen as reinforcing the need to suppress alternative goals to drive patients down the abstinence channel – or the opposing need make those alternatives more widely available to attract more problem drinkers into treatment. The abstinence camp would see only a complete break from drinking as a success and incremental gains as failures presaging relapse, while those same gains could look like success to commentators prepared to laud moderation-based recovery. For Roizen, “So distant and mutually unintelligible were the orientations of these two approaches that they could raise the specters of bad faith and fraud between the camps.”
The Sobells themselves were not above some re-interpretation of results consistent with a desire to present their controlled-drinking participants in the best light. In both their first- and second-year (free source at time of writing) follow-up reports, days spent in hospital due to drinking were presented as a negative outcome which meant the patient was not “functioning well” and had been admitted for “alcohol-related health problems (usually for detoxification)”, implying serious and prolonged relapse. Faced with their critics’ stress on hospitalisations as a signal of controlled drinking’s failure, and first-year data showing these events were more common among patients trained in controlled drinking, the Sobells re-interpreted hospital admission as a strategy for preventing serious relapse, in line with their programme’s “encouragement to intervene early when drinking got out of hand”. Their reasoning relied on the short duration of most admissions, but the same reasoning was not applied to comparison patients, and it was unclear whether short stays were actually therapeutic or (for example) devoted to patching up alcohol-related injuries. This ambiguity should not, however, obscure the fact that the comparison patients ‘compensated’ for their lower first-year hospitalisation tally with more days of alcohol-related imprisonment, surely an unambiguously bad outcome.
1. Personal communication from Ray Hodgson, December 2020.
2. “The hypothesis of the proposed project is that individualized behavior therapy for controlled drinking has not produced the effects reported by the investigators.” (Pendery M., Maltzman I. and Digan M. Follow-up study of effects of controlled-drinking treatment for alcoholics. Patton State Hospital research protocol, 5 May 1973. Quoted in the Sobells’ reply at http://dx.doi.org/10.1016/0005-7967(84)90084-6.
3. “First, the Sobells acknowledged the problem of interpretation when control subjects, although directed toward abstinence, were aware that controlled drinking was considered a potentially attainable goal and that they had been selected as appropriate subjects for that goal. [Editor’s note: from the available sources it is not clear that patients knew whether they had been “selected as appropriate subjects for that goal”. A 1973 report by the Sobells (https://doi.org/10.1016/S0005-7894(73)80074-7) says this decision was not made in front of the patients by their interviewer, but “After a majority staff decision”.] Second, the available data suggest that the experimental and control groups may have differed before they were treated. For instance, most of the controlled drinking subjects were admitted to Patton State Hospital earlier than most of the abstinence subjects. Thus, even were group comparisons appropriate, in our view they could not be made with confidence.”
4. Though the Sobells also said this was lower than might have been estimated: “We should note that the criteria used by the Committee for determining that a contact had occurred were somewhat conservative. For example, when a S was incarcerated for longer than a month and the agency was repeatedly contacted to verify the S’s presence, the Committee considered that only one contact had occurred.”
5. Or the 26 which can be calculated from a four-week interval over two years.
6. From http://dx.doi.org/http://dx.doi.org/10.1142/6579: “I accused the Sobells of fraud because the majority of patients we interviewed informed us that they were not interviewed every 3–4 weeks for 2 full years.”
7. None of the original reports are readily available and were not consulted for this essay.
8. Specifically the adequacy of the description of the their follow-up methods and the accuracy of their accounts of the frequency of follow-up contacts.
9. From https://doi.org/10.15288/jsa.1985.46.433: “Those who have commented on this controversy, including the Sobells‘ own defense, have established in every case that there is no essential difference between what the Sobells [and Caddy et al. (1978)] reported across the first 3 years and what Pendery et al. reported across those 3 years.” The cited article’s (https://doi.org/10.15288/jsa.1985.46.433) own analysis adds just “three more successes” to the Pendery data from the Sobells’ version of events.
10. “Nor, more interestingly, did the Committee discover any specific piece of data that was reported by Pendery et al. that was not also included in the Sobells’ published data … Since the Committee could then presume a one-to-one correspondence between the report in the Pendery et al. article and the actual hospital records, what was left to the Committee was the job of checking whether each of the hospitalizations in Table 2 of the Pendery et al. article was in the Sobells’ original data and then whether these data noted on data sheets actually match the Sobells’ published data … We did this for each person and found that the Pendery et al. and the Sobells’ data coincided.”
Next episode: “Evidence accumulates and expert opinion converges.” Moving from history to contemporary evidence, the heat dies down as it becomes clearer that neither controlled drinking nor abstinence has a definitive advantage as a treatment goal, and as evidence emerges about what types of patients do better with either strategy.
Full essay available on the Effectiveness Bank.
Email the author.