Side Effect Patterns in a Crossover Trial of Statin, Placebo, and No Treatment

Background Most people who begin statins abandon them, most commonly because of side effects. Objectives The purpose of this study was to assess daily symptom scores on statin, placebo, and no treatment in participants who had abandoned statins. Methods Participants received 12 1-month medication bottles, 4 containing atorvastatin 20 mg, 4 placebo, and 4 empty. We measured daily symptom intensity for each using an app (scale 1-100). We also measured the “nocebo” ratio: the ratio of symptoms induced by taking statin that was also induced by taking placebo. Results A total of 60 participants were randomized and 49 completed the 12-month protocol. Mean symptom score was 8.0 (95% CI: 4.7-11.3) in no-tablet months. It was higher in statin months (16.3; 95% CI: 13.0-19.6; P < 0.001), but also in placebo months (15.4; 95% CI: 12.1-18.7; P < 0.001), with no difference between the 2 (P = 0.388). The corresponding nocebo ratio was 0.90. In the individual-patient daily data, neither symptom intensity on starting (OR: 1.02; 95% CI: 0.98-1.06; P = 0.28) nor extent of symptom relief on stopping (OR: 1.01; 95% CI: 0.98-1.05; P = 0.48) distinguished between statin and placebo. Stopping was no more frequent for statin than placebo (P = 0.173), and subsequent symptom relief was similar between statin and placebo. At 6 months after the trial, 30 of 60 (50%) participants were back taking statins. Conclusions The majority of symptoms caused by statin tablets were nocebo. Clinicians should not interpret symptom intensity or timing of symptom onset or offset (on starting or stopping statin tablets) as indicating pharmacological causation, because the pattern is identical for placebo. (Self-Assessment Method for Statin Side-effects Or Nocebo [SAMSON]; NCT02668016)

is responsible. If symptoms disappear and reappear on stopping and restarting, does this prove the statin is the cause?
The SAMSON (Self-Assessment Method for Statin Side-effects Or Nocebo) trial enrolled participants with side effects sufficiently severe to have abandoned statin therapy. Each underwent a protocol of 12 statin, placebo, and no-tablet months. Daily symptom scores were collected using a smartphone app.
In this full report, we show every patient's full data. We explore the patterns of symptom onset and offset and discuss their diagnostic value.  Completed at least one month of trial for symptom burden calculation (n = 60) Only completed 1 -11 months (n = 11) Patient decision to discontinue intervention (n = 7) • Severe symptoms (n = 5) • Decided side-effects not statin mediated (n = 1) • Intercurrent social stresses (n = 1) Study team decision to discontinue intervention (n = 4) • Repeated failure to provide symptom scores (n = 2) • Other health issues preventing continuing with protocol (n = 1) Lost to follow-up (n = 0) Patient flow through the trial. Intensity, Onset, and Offset of Statin Tablet Symptoms S E P T E M B E R 2 1 , 2 0 2 1 : 1 2 1 0 -1 2 2 2

METHODS
The primary endpoint requested by our patient advisory group had the "nocebo ratio" defined as follows: The original intention was to calculate this ratio individually for each participant. However, some patients had extreme or indeterminable ratios, caused by the denominator of the ratio approaching or equaling zero, making the nocebo ratios very far from normally distributed. An independent statistician therefore recommended pooling the participants before calculating this nocebo ratio.
We therefore now report the marginal means of symptom intensity during statin, placebo, and no-tablet months, which remain valid for statistical analysis and hypothesis testing, and alongside this we report the resulting nocebo ratio. STATISTICAL ANALYSIS. We modeled change and variation in symptom intensity during and between statin, placebo, and no-tablet months, using mixed (multilevel) linear models. A sequence of 4 models was run. First, an "unconditional" model, which simply separated symptom intensity variance into within-month, between-month, and between-subject components, provided a baseline for subsequent model comparison. Second, we added a fixed effect of treatment to explain variance between months.
Third, we allowed the effect of treatment to vary between subjects. Finally, we fine-tuned the model by adding an autoregressive (lag) effect within subjects. The effect size and statistical significance of the treatment effect, and post hoc tests between the marginal means for each treatment type, provided a test of our primary hypothesis, namely whether symptom intensity would differ significantly between the 3 types of months.
As there were no prior data on the relative size of nocebo and statin side effects, we followed the advice of Maas and Hox to have a sample of at least 50 participants to help ensure unbiased estimates of the treatment-level SEs (12).
For a month of data to be included in the numerical analysis, the patient had to have reported at least 10 symptom scores. Participants who had to stop their tablets early because of intolerable symptoms but continued submitting symptom scores were not excluded from the analysis, but their symptom scores following tablet cessation were not used. Within participants who satisfied these inclusion criteria, any further such attritional missing data were handled using multiple imputation by chained equations (used for the primary analysis) (11). The  Values are mean AE SD, n (%), n, or median (interquartile range), unless otherwise indicated.
Nocebo ratio ¼ Symptom intensity on placebo À Symptom intensity on no tablets Symptom intensity on statins À Symptom intensity on no tablets To assess whether the severity of symptoms or the  Table 1.

TIME-COURSE OF ONSET AND OFFSET OF SYMPTOMS.
We studied whether the intensity of symptom onset when starting tablets indicated whether that tablet was statin (Figure 3). The top panel of Figure 3 R 2 1 , 2 0 2 1 : 1 2 1 0 -1 2 2 2 they reveal several patterns. Figure 2 shows 6 exemplar patients whose symptom scores we will now cover in more detail. Some participants (eg, participant 53) essentially experienced no side effects during the trial, despite intolerable side effects with statins previously. Others (eg, participant 10) had intense symptoms throughout, regardless of whether they were taking tablets or not. First, in patients with symptoms during both statin and placebo months, we would not be able to discriminate between the nocebo effect and everpresent background symptoms. Second, a theoretical concern might be that SAMSON did not wait long enough for the statin side effects to wash out, and these were causing the high placebo scores. The inclusion of no treatment periods mitigates this concern because the no treatment months had significantly fewer symptoms (P < 0.001) than the placebo months. SAMSON tested for this by having both no-tablet and placebo tablet arms, which reveals the expectation of side effects (ie, nocebo) to be the dominant contributor, because symptoms were much worse on placebo tablets than no tablets (P < 0.001), and no different between placebo and statin (P ¼ 0.388). Furthermore, 10 participants who had to stop tablet months early managed, after discussion of their personalized results, to successfully restart statins.

DANGER OF CURRENT PRACTICE OF INFORMAL
EXPERIMENTS. Prompt onset and offset of symptoms after starting and stopping tablets is often interpreted by patients and clinicians as evidence of causation.
Our data indicate that this is true, but because the kinetics are identical between placebo tablets and statin tablets (Figures 3 and 4), the causation is from taking a tablet, rather than from the tablet being a statin.
The danger of the informal experimentation that is currently recommended in North America (19), Europe (20), and the United Kingdom (21) is that even if there is a preplanned schedule and daily symptom scoring, without a placebo it is inevitable that nocebo   Clayton, and Dr Neil Chapman, who have provided independent, expert professional advice throughout the study on patient safety data. Finally, the authors thank Juliet Holmes, who has been key to coordinating the trial, and Yousif Ahmad, who greatly assisted them in preparing the manuscript.

FUNDING SUPPORT AND AUTHOR DISCLOSURES
This study was funded by the British Heart Foundation (PG/15/7/ 31235), which had no role in study design, data collection, data analysis, data interpretation, or writing of the report. This study was supported by the National Institute for Health Research Im-