There is a reasonable mechanism by which persistent viral infections might raise the risk of Alzheimer's disease: amyloid-β is an antimicrobial peptide, a part of the innate immune system. The presence of viral particles will contribute to greater production of amyloid-β, which will accelerate the pace at which amyloid-β might aggregate in older individuals due to an imbalance between production and clearance. The aggregates then cause the usual progression to neural inflammation, damage, and cognitive decline. Does the epidemiological data support a role for persistent herpes viruses in Alzheimer's risk, however? Previous studies suggested yes, but here researchers dismantle and refute one of those studies, while suggesting that people need to be more careful when using statistics. This sequence of events happens more often than you might think in the research community.
Like all types of dementia, Alzheimer's disease is characterized by massive death of brain cells, the neurons. Identifying the reason why neurons begin and continue to die in the brains of Alzheimer's disease patients is an active area of research. One theory that has gained traction in the past year is that certain microbial infections, such as those caused by viruses, can trigger Alzheimer's disease. A 2018 study reported increased levels of human herpesvirus in the postmortem brain tissues of more than 1,000 patients with Alzheimer's disease when compared to the brain tissues of healthy-aging subjects or those suffering from a different neurodegenerative condition.
Surprisingly, when researchers reanalyzed the data sets from the 2018 study using the identical statistical methods with rigorous filtering, as well as four commonly used statistical tools, they were unable to produce the same results. The team was motivated to reanalyze the data from the previous study because they observed that while the p-values (a statistical parameter that predicts the probability of obtaining the observed results of a test, assuming that other conditions are correct) were highly significant, they were being ascribed to data in which the differences were not visually appreciable. Moreover, the p-values did not fit with simple logistic regression - a statistical analysis that predicts the outcome of the data as one of two defined states. In fact, after several types of rigorous statistical tests, they found no link between the abundance of herpes viral DNA or RNA and likelihood of Alzheimer's disease in this cohort.
"As high-throughput 'omics' technologies, which include those for genomics, proteomics, metabolomics and others, become affordable and easily available, there is a rising trend toward 'big data' in basic biomedical research. In these situations, given the massive amounts of data that have to be mined and extracted in a short time, researchers may be tempted to rely solely on p-values to interpret results and arrive at conclusions. Our study highlights one of the potential pitfalls of over-reliance on p-values. While p-values are a very valuable statistical parameter, they cannot be used as a stand-alone measure of statistical correlation - data sets from high-throughput procedures still need to be carefully plotted to visualize the spread of the data. Data sets also have to be used in conjunction with accurately calculated p-values to make gene-disease associations that are statistically correct and biologically meaningful."