On 7/10/2020, I discussed a study suggesting that homeopathy improves the quality of life and survival of cancer patients. Now, these data have been carefully scrutinized by a group of members of the „INH“ and „Initiative für Wissenschaftliche Medizin“.
By guest bloggers Norbert Aust and Viktor Weisshäupl
The first impression of the results of the study on the adjunctive homeopathic treatment of patients with non-small cell lung cancer (NSCLC) is that of a seemingly rigorous trial with valid results. But a more thorough review yields different insights:
- The methods and definitions were pre-determined in a protocol and seem to have been maintained up to the end. But the date given in the document pointing at some point in time before enrollment began is wrong and misleading: This protocol was first published by uploading it to the register only two months after data assessment was completed with outcomes presumably available.
- The data initially saved to the register are not in agreement with the information given in the published paper: important definitions were subjected to considerable modifications while the study was underway. None of these modifications are mentioned in the paper, neither a rationale nor a comment of their impact on the results was provided.
- Some of the modifications with presumably heavy impact on the results were introduced with the upload of the protocol only, that is two months after data collection was completed. These were (a) a massive extension of the exclusion criteria: the number increased from 1 during initial registration to 20 in the final paper. and (b) an equally massive reduction of the follow-up time for the primary endpoint from two years to 18 weeks.
- The paper discloses no reason why the additional exclusion criteria were introduced. Their selection seems arbitrary without any apparent necessity arising from the trial itself.
- The patients who did not meet the added criteria and were thus excluded are not mentioned in the publication. The CONSORT flow chart does not give information either of their number or of the point in time when they were excluded.
- The survival curves of the placebo and verum groups show some aspects that arise if the inter-group difference was due to the exclusion of unfavorable data.
- It is hard to imagine that, in this trial, the homeopathic preparations had strong effects on the patients’ health, while other rigorous studies or systematic reviews failed to notice such effects.
Altogether, it seems much more plausible to assume that the positive results were achieved by post hoc data manipulation, namely by omitting patients with unfavorable outcomes, than by rigorous and valid science. A retraction of the paper seems the only appropriate measure to avoid misleading the public.
Due to its outstanding results, the study about adjunct homeopathic treatment of non-small cell lung cancer patients was met among homeopaths with enthusiasm. However, in this article, we will show that the enthusiasm is unjustified because the results may not be based on a rigorous trial meeting established scientific criteria. Crucial definitions were modified, while the study was underway or even after data collection was completed. It stands to reason that this introduced bias in favor of homeopathy.
For this analysis, we considered the following sources of information :
- the text of the published paper (link)
- the data that were uploaded during registration (link)
- the history of changes of the registered data (link)
- the study protocol included in the registration (link)
As all of this information is readily available on the internet, it is easy to double-check our findings and verify our statements. We also submitted a letter to the editor of the Oncologist, the journal where the paper was published which has not yet been published (status 06-06-2021).
At first glance, the study meets the requirements for reliable evidence.
- There is a study protocol dated January 11, 2011, well before recruiting of participants started. It provides definitions that were used until the end.
- The study was registered at ClinicalTrials.gov before recruiting started.
- The methods of randomization and blinding are suitable to meet the requirements for a low risk of bias rating.
- The presentation of the paper follows the principles set out in the CONSORT-Statement.
- The paper was published in a peer-reviewed journal of some reputation.
The study yielded formidable results in favor of homeopathy: In the group that received the adjunctive homeopathic treatment, the quality of life improved continuously throughout the follow-up time, while the patients in the placebo group deteriorated. In addition, the median survival time was only about two-thirds compared to the patients in the homeopathy group. However, the impression of a valid study does not stand up to closer scrutiny when the history of changes is taken into account.
Changes in study parameters
Between the initial registration and data upload in January 2012 (Link), shortly before recruiting started in February 2012, and the publication in October 2020, multiple changes in essential study parameters occurred:
|Registration January 2012||Publication October 2020|
|Number of participants||600||150|
|Number of study arms||2||3|
|Number of exclusion criteria||1||20|
|Follow-up time for Quality of life||104 weeks (*)||18 weeks|
|Number of cancer types||3||1|
|(*) Derived from “Time Frame: 7 Years” minus the recruitment period of 5 years.|
Note the drastic reduction in the follow-up time for quality of life by more than 80 % which was defined as the primary endpoint. Furthermore, note the substantial increase in the number of exclusion criteria. Both issues will be discussed in more detail below.
In contrast to the requirements for a rigorous and valid trial, these modifications are not mentioned in the published paper, and no rationale is given as to why they became necessary. As a consequence, the authors do not discuss the possible impact these modifications may have had on the results.
The study protocol
A study protocol is available in the registration database. It was first uploaded on September 18, 2019, about two months after the end of data collection in July 2019 (Link). The document itself is dated January 11, 2011, which would place it about a year before the study was registered. However, this date is obviously wrong: there are substantial discrepancies between the parameters specified in the protocol and the data provided one year later during initial registration:
|Protocol, allegedly January 2011||Registration January 2012|
|Number of participants||300||600|
|Number of study arms||3||2|
|Number of exclusion criteria||9||1|
|Follow-up time for Quality of life||18 weeks||104 weeks (*)|
|Number of cancer types||1||3|
|(*) Derived from “Time Frame: 7 Years” minus the recruitment period of 5 years.|
We see no sensible explanation why the parameters given in the study protocol allegedly compiled in January 2011 are in line with the publication nine years later, but not with the registration only one year after the protocol was compiled. The only sensible conclusion seems to be that this protocol was not completed on the date indicated, but at a much later point in time, maybe just shortly before its upload (September 2019). This impression is corroborated by the information presented in the document that was not available on the date given: On page 10 the software package used in data analysis is referenced as “IBM SPSS statistics 25.0” while, at the beginning of 2011, when the protocol was allegedly compiled, the current version number of this package was 19 only.
A second clue: Also on page 10 there is a reference “(EORTC-QLQ-C30 remaining dimensions; SF-36; subjective well-being)25.” with the number 25 indicating some reference. And some references that is, but not in the protocol – this does not have any references – but in the published paper, where the 25 indicates a paper on the SF-36 questionnaire. So it stands to reason that the number in the protocol originates from some messed up copy and paste procedure from the draft of the paper. Which would indicate that the paper and the protocol were at least partially developed in parallel.
It seems therefore reasonable to assume, that the protocol was finished only shortly before it being uploaded in September 2019, that is two months after data collection was completed.
However, the obviously inaccurate date given in the protocol supports the impression that the study parameters were set a year before the study began and were consistently maintained during the course of the trial, which is not the case, as the above tables show.
Change in exclusion criteria
The initial registration data list pregnancy as the only exclusion criterion. But with the upload of the protocol, which took place two months after data collection was completed, the number of exclusion criteria was increased to nine, only to be enlarged once again in the final publication to the final number of twenty. It is beyond any doubt that at least the final increase of eleven criteria took place after the data collected from the patients were available. But all this is neither disclosed in the final paper nor is there any rationale given for this action.
The patients excluded by the additional criteria never appear anywhere, they are not included in the CONSORT-flow-chart, Fig 3 in the study. It is obvious that some patients were excluded: What was the reason to define such an abundance of criteria, if they were not to be applied? As a consequence, the CONSORT diagram seems to be incomplete which would be in violation of the CONSORT statement.
Thus, an unknown number of patients seems to have been excluded from the study by criteria defined at a time after data collection was completed with outcomes available. After all, eleven of the exclusion criteria were established even after the protocol had been uploaded, at least those were established well after the patients’ results were available.
This raises the question of why these exclusion criteria were introduced. One would assume that an intervention to treat stage III and stage IV lung cancer patients should be effective under the conditions that are usually present in such patients. One would expect that patients somewhat advanced in age, like in this study, usually suffer from some health problems, regardless of their cancer condition. What is the sense of excluding patients with hematological, hepatic, or renal pathology, with coronary heart disease or rheumatism? Homeopathy is claimed to be able to treat comorbidities based on the assessment of symptoms independent of what disease they belong to. And this apparently was the idea at the start of the trial where only pregnancy was specified as an exclusion criterion, while it was understood that elderly patients to be enrolled in the study would suffer from some additional medical problems.
On the other hand, not all health conditions that are associated with advanced age were excluded. Diabetes, hypertension, gastrointestinal diseases, or COPD were no reason to exclude any patient from participation. Only very few of the criteria are somewhat self-explanatory as to why they were defined as exclusion criteria, e.g. if a patient was unwilling to give her informed consent.
Altogether, the assumption seems reasonable that more patients had participated in the trial than accounted for in the publication, and that an unknown number of them were excluded according to criteria that were not present until after data collection was completed. If so, a substantial bias was introduced.
Median survival time
Here, we will focus on the comparison between the homeopathy and placebo groups and leave aside the third group not receiving any additional treatment at all.
If the favorable result in survival really was established by dropping unfavorable data, this might be recognized in some characteristics of the survival curves. Therefore, we modeled this situation starting with two random distributions somewhat tweaked to resemble the typical shape of natural survival functions.
This graph shows the two distributions (n = 80) defined in the range of 0 to 200 as thin lines. Both are very similar to each other with median survival at 27 weeks. If 15 of the 20 patients with the shortest survival are dropped from the thin blue line this would result in the solid blue line (“Hom”). If, on the other hand, 15 out of 20 patients with the longest survival are dropped from the other distribution this would yield the solid red line (“Plac”).
The new functions show some characteristic properties:
- In the red line, median survival drops by 8 weeks to 19 weeks.
- In the blue line median survival rises by 12 Weeks to 39 weeks.
- The difference between the two functions arises from of the first 12 weeks alone. With the blue line, 8 people died, with the red line 23 people died during the first 12 weeks. After week 12 up to week 80, the same number of fatalities occur in both groups (blue: 36, red: 37).
After week 80, the two functions start to converge, which is due to the fact that at some future point all the patients of both groups will be dead. The survival functions that are reported in the study show the same characteristics.
Assuming that homeopathy did not have any effect, both groups should show more or less identical survival functions. In the paper 10.1 months = 303 days is cited from literature as to be expected under conventional care, maybe with some margin to the better because the data that yielded this value of 10.1 months were more than six years old at the time of the trial. The survival functions allegedly found in the trial show:
- Median survival time with the placebo group is reduced by 46 days compared to conventional care alone.
- Median survival time with the homeopathy group is increased by 132 days compared to conventional care alone.
- The advantage of homeopathy arises within the first 9 weeks alone, where only two patients died (out of 51) in the homeopathy group compared to 11 (out of 47) in the placebo group. After this initial phase, the groups developed in parallel: By the end of the two-year follow-up time an additional 26 patients died with homeopathy, and about the same, namely 25 patients died with placebo.
The inevitable convergence of both functions apparently started outside the two-year follow-up. In other words, the survival functions given in the study for placebo and homeopathy treatments show characteristics that match what you would be expected, if two very similar functions were manipulated by dropping unwanted results, i.e. “good” survival data from placebo and “bad” survival data from homeopathy functions. After week 9, the two functions develop parallel to each other, indicating a lack of effect of homeopathy even though the treatment continued until the death of the patient or the end of the study. However, with ongoing effective treatment, the functions should continue to diverge. It seems implausible that homeopathy should be effective on a short time basis only, with a sudden complete loss of effectiveness later on.
Reduction of observation time
Quality of life was defined as the primary endpoint. On initial registration, it was specified that patients should be observed for the entire seven-year duration of the study which, allowing for the recruitment period of five years, results in a follow-up time of two years or 104 weeks for each individual patient. According to the information provided in the study, this was indeed done: “Patients were followed up every nine weeks until death” (or until the end of the study, of course), and questionnaires were completed to determine the quality of life.
The reduction of follow-up time from 104 to 18 weeks was first introduced when the protocol was uploaded. So it is obvious that this substantial reduction occurred after data collection was completed and that data from more than 80 % of the originally defined follow-up were omitted.
Incomplete outcome reporting, especially when a larger scope was defined at the beginning of the study, is considered a source of substantial bias and a major shortcoming in clinical trials: Maybe patients initially experienced an improvement in their quality of life due to whatever effect – but what were the results after this initial phase? Why were they omitted? Perhaps because they got worse than in the placebo group? The long-term development would have been a vital aspect for the evaluation of efficacy – and the study originally was designed to evaluate such long-term effects. Yet, the authors’ conclusions on the quality of life – notably: the primary outcome criterion – are based on less than 20 % of the follow-up in which a positive effect may have occurred due to bias or by chance. To extrapolate from this short time to the total period is not justified and may be misleading. A detailed review of the quality of life results is meaningless: they do not disclose any long-term effects and they are subject to bias caused by the post hoc exclusion of patients anyway.
The overall evidence on the effectiveness of homeopathy is not encouraging. The quintessence of all systematic reviews that have looked at homeopathy as a whole is that some marginal effect may be found, if all studies are included in the review, regardless of their quality. But this result is questionable due to the generally low quality of the primary studies (Link, in German). However, when quality is taken into account, the systematic reviews do not produce robust evidence for any positive effect beyond placebo. In addition, no review could identify a single condition in which homeopathy is of well-established therapeutic benefit.
This study on NSCLC contradicts the long-established and often-confirmed evidence. During the follow-up time for the patients who actually received the prescribed homeopathic preparations, the quality of life improved steadily in all subscales – even down to the patients’ financial situation – whereas the opposite was observed in the placebo patients. In addition, the mean survival time was about two-thirds longer for the homeopathy patients than for the placebo group.
After 200 years of clinical research into homeopathy, it seems unlikely that such a powerful effect of homeopathy should not have been noticed before. Another scenario seems to be much more plausible:
- The survival times of the placebo group were worse than the data from the literature. This could be due to the fact that patients with relatively good outcomes were excluded by the introduction of additional exclusion criteria.
- The survival times of the homeopathy group were considerably better than expected. This could also be due to the additional exclusion criteria, in that patients with poor outcomes were excluded retrospectively.
- The long time frame where the survival functions run in parallel from week 9 onwards until the end of the two years observation period indicates the lack of effect of the homeopathic treatment. The advantage occurring in the first nine weeks alone seems to be the result of unwanted data being dropped.
- In the case of quality of life (after all, not a “hard” criterion, but based on information from the patients ), the advantage in survival would have initially created a positive effect for the homeopathy group. Then, reporting was discontinued, once the initial positive effects presumably caused by the selective omission of patients had ended.
In conclusion, it seems likely that the substantial modifications of crucial study parameters that occurred after the study had been started and results had become available biased the results in favor of homeopathy. Therefore, this study does not meet strict scientific standards that were established to exclude any confounding factors or biases. If our analysis is correct, the results of this study are invalid, and the authors’ conclusions are not justified. Retraction of this study seems to be appropriate.
Reference Frass M, Lechleitner P, Gründling C et al. Homeopathic Treatment as an Add-On Therapy May Improve Quality of Life and Prolong Survival in Patients with Non-Small Cell Lung Cancer: A Prospective, Randomized, Placebo-Controlled, Double-Blind, Three-Arm, Multicenter Study. The Oncologist 2020;25:e1930–e1955 https://theoncologist.onlinelibrary.wiley.com/doi/epdf/10.1002/onco.13548