Table 1 summarizes the results for all studies, for peer-reviewed studies, after exclusions, and for specific outcomes. Fig. 2 plots individual results by treatment stage. Fig. 3, 4, 5, 6, 7, 8, 9, 10, 11, and 12 show forest plots for random-effects meta-analysis of all studies with pooled effects, mortality results, ventilation, ICU admission, hospitalization, progression, recovery, cases, peer reviewed studies, and long COVID.

To avoid bias in the selection of studies, we analyze all non-retracted studies. Here we show the results after excluding studies with major issues likely to alter results, non-standard studies, and studies where very minimal detail is currently available. Our bias evaluation is based on analysis of each study and identifying when there is a significant chance that limitations will substantially change the outcome of the study. We believe this can be more valuable than checklist-based approaches such as Cochrane GRADE, which can be easily influenced by potential bias, may ignore or underemphasize serious issues not captured in the checklists, and may overemphasize issues unlikely to alter outcomes in specific cases (for example certain specifics of randomization with a very large effect size and well-matched baseline characteristics).

The studies excluded are as below. Fig. 13 shows a forest plot for random-effects meta-analysis of all studies after exclusions.

Exercise is not expected to have a direct antiviral effect; rather, any benefits are hypothesized to stem from improving immune system function (host-directed efficacy). For such immunomodulatory treatments, we may expect a gradient of efficacy across outcomes, with minimal impact on preventing initial infection (PCR+ cases), but increasing efficacy as the disease progresses to more severe stages.

For exercise, random-effects meta-regression shows a significant association with increasing efficacy for more severe outcomes. For every one-step increase in severity (cases → hospitalization → critical outcomes: ICU, ventilation, and mortality), meta-regression shows that the Risk Ratio decreases by a factor of 1.22 (slope β = -0.20 [-0.32 to -0.09]; p = 0.00059). In other words, the protective effect becomes stronger as the outcome becomes more critical.

The trend across outcomes is consistent with a host-directed immunomodulatory mechanism rather than a direct antiviral effect. This reinforces the reliability of the overall finding that exercise reduces risk for COVID-19. If the observed efficacy was due to a systematic bias increasing the efficacy of outcomes, it would not be expected to create a trend across outcomes that matches the biological mechanism. Moreover, systematic biases are more likely to effect subjective outcomes than the objective mortality outcome. The fact that exercise shows the strongest effect on the most objective outcome (mortality) argues against such bias explaining the results.

With 68 studies for exercise, we can analyze trends across studies. A significant interaction that matches the biological mechanisms provides further confirmation of efficacy. Analysis of the efficacy gradient across outcomes shows a significant interaction, with p = 0.00059.

This reinforces the reliability of the overall finding that exercise reduces risk for COVID-19. If the observed efficacy was due to a systematic bias increasing efficacy in results, we would not expect analysis across studies to find this significant interaction that matches the biological mechanisms.

Low-cost treatments were subject to bias and censorship during the pandemic. Scientific bias is seen in the design, analysis, presentation, and selective reporting of studies, which often favored negative results. A similar bias is seen in the media coverage for low-cost treatments. While broadly seen, bias was particularly notable for ivermectin and hydroxychloroquine, e.g., Scott Alexander noted that "if you say anything in favor of ivermectin you will be cast out of civilization and thrown into the circle of social hell reserved for Klan members and 1/6 insurrectionists. All the health officials in the world will shout 'horse dewormer!' at you and compare you to Josef Mengele."13.

We analyze media coverage for the 227 treatments we cover using Altmetric31 and PlumX Metrics32. Altmetric covers ~14,000 news outlets33 and PlumX covers ~55,000 news sources34. The two services are complementary with coverage of different news sources and different methods for identifying articles not containing a DOI. We consider studies to have received significant media coverage if they were covered by at least 100 articles in either Altmetric of PlumX34,35. The most widespread study was covered by 1,047 outlets. Studies covered by major syndicated sources shows hundreds of news articles, while studies covered by few outlets typically appear in niche medical news sources or press releases. Fig. 14 and 15 show bias towards negative results for low-cost treatments, in contrast to the opposite bias for high-profit treatments.

Fig. 16 shows that the mainstream media covered only a small fraction of the positive studies for exercise.

The media bias may result in widespread incorrect perceptions on the relative efficacy of high-profit and low-cost treatments. The impact is significant—increased cost limits the use of high-profit treatments and treatment equity. High-profit treatments were also more difficult to access, especially for earlier treatment which improves efficacy and minimizes community transmission.