August 6, 2025

Clinical Insights Into Observational Studies

By Dr. William Ondo, Director of the RLS Quality Care Center at Houston Methodist Neurological Institute

Medical epidemiological studies, also called observational studies, assess information from existing data sources. These studies often make news headlines, reporting “associations” between some factor (e.g., medicine, food, activity) and some outcome (typically a feared medical condition). No epidemiology/observational study can ever deduce that one factor “caused” the other, a distinction that is often blurred by media sources. A classic example: there is an extremely high association between ice cream consumption and deaths by drowning. However, it is unlikely that ice cream consumption causes death by drowning and unlikely that death by drowning causes more ice cream consumption. It is much more likely that both occur more in summer, which explains the association. Although a useful, fast, and inexpensive mode of study has heralded many medical advances, in isolation, retrospective epidemiology studies are among the least scientific and potentially misleading forms of analysis.

One recent observational study correlated the use of gabapentin with developing dementia.¹ This study incorporated many of the statistical strategies used to improve the accuracy of observational studies. However, even the most strenuous and well-meaning analysis can’t overcome the potential for bias in any retrospective (analyzing previously collected data) study. This specific study compared 26,416 patients who used gabapentin for low back pain against 1,103,678 patients who did not take gabapentin for low back pain. All diagnoses were based on recorded standardized electrical diagnosis codes. The researchers from this study then used a complex process to make the larger non-gabapentin group more similar to the gabapentin group by specifically selecting a subset of 26,000 individuals from the non-gabapentin group to use in the direct comparison. By doing this, the researchers can “correct” for some intrinsic differences to make the comparison relatively more fair. Patients with other established risk factors for dementia that might bias the group (independent of gabapentin use) were excluded. At the end of this correction, the two groups were similar for age, sex, race, hypertension, diabetes, heart disease and some other disorders that have been associated with developing dementia, based on electronic medical record data.¹

However, researchers can’t correct for every factor. Fortunately, the authors of this study included the baseline characteristics of the 26,000 gabapentin group vs. the 1.1 million non-gabapentin group. Overall, the gabapentin group was much “sicker,” older, and had many additional factors that would otherwise increase their risk for dementia. Although the researchers corrected for some of these, they could not correct for many other potential dementia risk factors, such as hearing loss, social isolation, obesity, physical activity/exercise, and general attitudes toward health/healthy lifestyles. The published results do not allow us to determine whether the gabapentin group had more of the unmeasured dementia risk factors. However, because they had higher risks across almost every other reported baseline characteristic, it is reasonable to infer that they likely had more of these unmeasured risk factors as well. It was also indicated that the gabapentin baseline group were prescribed a greater number of opioids and other pain treatments. Based on this information, the gabapentin baseline group likely had more severe pain issues, which would at least suggest less exercise and more social isolation.

Potentially the biggest uncorrected risk factor for these types of observational studies is the frequency of health care visits. The more a person seeks medical assistance, the more likely they are to receive treatment and obtain a diagnosis for medical conditions. No individual who has not seen a physician is diagnosed with dementia and patients who visit a physician less often are also less likely to receive a subsequent diagnosis of dementia. This is a major, although underrecognized, intrinsic bias with all observational studies.

Other issues that may account for bias include the accuracy of the medical records, since no patient was interviewed or completed any type of questionnaire for this study. Further, generalizations to one disease (back pain) cannot always be attributed to another (RLS). For example, alpha-2-delta ligand medications (gabapentin, gabapentin enacarbil and pregabalin) improve sleep in RLS, which is known to lessen dementia risk in general. Specifically, they increase deep (delta) sleep, which is also hypothesized to reduce dementia.

Unfortunately, proper scientific studies to prove causality require prospective (collecting data moving forward in time) trials, which are lengthy, expensive and very challenging. With respect to this study, proving an association between gabapentin and dementia would require randomizing thousands of similar people to take either gabapentin or a placebo, with no chance of taking similar medicines. Participants are followed regularly for 10-15 years, examining how many in each group develop dementia. That idealized version of the study is essentially impossible on practical, financial and ethical grounds. There are high quality, but short (12-48 week) randomized prospective trials demonstrating the benefit of gabapentin enacarbil and pregabalin for RLS, however these can’t definitively eliminate the possibility of side effects that occur only after many years. Interestingly, there have been only a few small studies of gabapentin for RLS, but because of its low cost, long-established safety profile, and similarity to gabapentin enacarbil and pregabalin, it is commonly used.

At this point, I do not feel this association study of gabapentin with dementia should deter people from using gabapentin for RLS. We consider this medicine to be among the safest overall options for RLS.

1. Eghrari NB, Yazji IH, Yavari B, Van Acker GM, Kim CH. Risk of dementia following gabapentin prescription in chronic low back pain patients. Reg Anesth Pain Med. Published online July 10, 2025. doi:10.1136/rapm-2025-106577