Fatigue in Cancer Treatment Studies: Analysis of Placebo Arms

Discussion

This study combined the data of placebo arms of randomized clinical trials of patients with cancer of any age, and determined the overall frequency of fatigue/asthenia to be 21.2%.

Placebo is the Latin word for “I will please”, describing the therapeutic effect of the physician’s expectation transmitted to the patient’s observation and reporting, and it is considered positive: the placebo effect typically reduces the reported symptom burden. The placebo effect is more likely to occur for AEs that mostly rely on the patient’s sensation, such as fatigue. However, in clinical trials, placebo is an actual medication, which might indeed have adverse effects, even though there is no active agent. Injections cause pain and possibly infection even when the drug itself does not. Niwa and colleagues described a case of severe fatigue caused by an oral medication that was thought to be an inactive placebo, but the capsule included caprylocaproyl polyoxylglyceride, now suspected to be the causative agent of the fatigue (11). The data provided here do not allow determination of whether a placebo reduces or increases symptoms. However, for the purpose of determining baseline data to judge observations made in drug-development studies, such findings are irrelevant. In any case, data derived from placebo arms are a better data source than those without placebo: Preparations with active agents also have placebo effects, and in the goal to determine the potential causality of the active agent, the placebo effect is an unwanted influencing factor. Using study data from placebo arms as comparator allows some degree of control over the undesired data effect.

Fatigue and asthenia have an overlapping meaning. The Medical Dictionary for Regulatory Activities MEDRA lists both as ‘preferred terms’ under the high-level term “Asthenic conditions”, the high-level group term “General system disorders” and the System Organ Class “General disorders and administration site conditions”. The word “fatigue” is more commonly used, and not just in medicine: In mechanical engineering it describes weakness of metals occurring after repeated mechanical stress. When describing symptoms of a patient with cancer, typically the patient might be described by both “fatigue” and “asthenia” or none of the terms. The overlapping meaning makes it unnecessary to use both in documentation or to analyze them separately when both terms had been used anyway, In fact, in aggregating AE data as percentage of patients, the use of two almost synonymous terms will result in inaccurate, low values for each of the terms. We found the two terms not to be correlated, which further confirms that investigators typically pick one of them. In summary, our findings create doubts in the wisdom of listing both words as ‘preferred terms’. For the purpose of using frequencies to infer drug causality in the context of oncology drug development, we recommended combining the two terms as one quantitative variable.

Fatigue/asthenia can be treated when a somatic cause such as the tumor itself, anemia, hypothyroidism or other hormone disorder is known. Treatment of fatigue/asthenia becomes more difficult in the absence of a treatable cause. Among patients with primary brain tumor, a literature analysis found insufficient evidence for any pharmacological or non-pharmacological treatments for fatigue (12). In contrast, fatigue in menopause was reported treatable with armodafinil (13), and among patients with cancer, exercise and behavioral therapy was reportedly effective (14). Similarly, cognitive behavioral therapy for insomnia was effective in cancer survivors (15). Our data support the concept of treating insomnia to address fatigue, since we found a correlation between the frequency of insomnia and grade 3 and higher fatigue. Furthermore, we found an even closer correlation between fatigue and nausea, which may prompt the hypothesis that antiemetics might also be effective against fatigue in the context of cancer.

Matching of influencing variables is a key component when comparing frequencies in aggregated data. Typically, the assumption is that a tumor response is dependent on the molecular profile of the tumor cells, while AEs are more dependent on demographic variables such as age and comorbidity (5). We found a relevant influence of cancer diagnosis on reporting of fatigue/asthenia. However, other demographic variables remained surprisingly insignificant. Of note, comorbidity may be considered the most influential covariant for AE frequencies, and fatigue might be influenced by various psychological factors. Unfortunately, it was not possible to extract data from the medical literature to address the hypothesis that comorbidities influence fatigue. Thus, the most appropriate benchmarks to be generated from these data may take cancer diagnosis into account, as far as sufficient data are available, but none of the other covariates can be included.

Drug causality is assessed in various workflows. Signal detection often relies on thresholds that can easily be implemented by automated processes monitoring large databases. Any frequency of fatigue/asthenia that is higher than the highest ever reported in placebo arms should be considered drug-related, and flagged. These benchmarks are: 83%, 19%, 17%, and 1% for grade ≥1, ≥2, ≥3, and ≥4 (Table III). Unfortunately, the ranges of reported frequencies were very large, limiting the usefulness of these upper limits of normal. A more sophisticated approach would be to use the actual numbers of patients reported in placebo arms (Table III) and compare them to the study observations, implementing exploratory chi-square tests or more advanced mathematical methods.

Various limitations of aggregating data from published AE tables have been observed over this series of analyses (5-7). Simpson’s paradox is the phenomenon in which a trend appears in several groups of data but disappears or reverses when the groups are combined (16). It can be avoided when confounding variables and causal relations are appropriately addressed. We found cancer diagnosis to be a significant covariant, enabling corrections for this variable. Other interesting observations were made which may also appear as limitations for certain interpretations. Most of these can be summarized as being the result of the information flow, starting with reporting of the AE by the patient, followed by grading and data entry by the local investigator, and after further steps ending in publishing customs: We showed more reporting of headache in healthier patients (5) and a lack of correlation between fatigue and asthenia. Both findings are likely caused by systematically unequal reporting: headache might not be worth mentioning in patients with cancer who have numerous, more severe, symptoms, and asthenia is not entered when fatigue is already documented. Moreover, the degree of reporting diligence varies widely – publications with thorough AE reporting have higher frequencies for all AEs, and combining these data with those generated less diligently will create artificially positive correlations, we even found correlations for opposing medical concepts, such diarrhea and constipation (7). None of these findings reflect actual biological or medical phenomena. However, they accurately reflect clinical study data given the current AE classification and data aggregation methods. Newly documented phase II study data are subject to the same environment. Therefore, when external controls are necessary, those should be derived from clinical trials to allow comparison of frequencies under the same conditions.