Heal House Vitiligo Research 58 · Review
Can scanning all diseases with a computer uncover hidden comorbidities of vitiligo?
Automated Mass Screening for Comorbidities of Vitiligo Using the National Health Insurance Database
Last reviewed
In short
Instead of looking only at preselected diseases, we used a computer to scan all diagnosis codes. In doing so, we found 45 conditions associated with vitiligo. Unexpected links, such as melanoma and obsessive-compulsive disorder, were also revealed.
Typically, when examining whether a certain disease is commonly comorbid with vitiligo, researchers preselect the diseases they suspect and then confirm them. This makes it easy to miss associations that no one suspected. So I turned the idea around. I wondered what would happen if, instead of a person choosing, a computer automatically scanned all diagnosis codes from start to finish. This study, which I led as corresponding author, used the entire National Health Insurance database in that way, attempting to find hidden conditions associated with vitiligo.
Q1 How exactly did the screening work?
We included 90,297 patients with vitiligo and 90,297 age- and sex-matched controls. Instead of a person, a computer automatically compared each of the 1,265 diagnosis codes one by one. Using National Health Insurance data from 2007 to 2017, we scanned virtually all disease classification codes from A to Z. By examining the entire set without preselecting assumptions, we could avoid missing unexpected associations. However, this automated scanning can introduce chance associations, so we also checked for potential bias.
Q2 What did the screening find?
We found 45 diseases with a significantly increased association with vitiligo and 6 diseases with a decreased association. This confirmed known links with autoimmune diseases, of course. It also newly highlighted less-known associations, such as melanoma (a type of skin cancer) and obsessive-compulsive disorder. Conversely, femoral fractures were less common in vitiligo. Representative results are summarized in the table below. However, this is a broad scan of 'possible associations,' so each one requires further confirmation.
Vitiligo-related signals revealed by automated screening
| Item | Content |
|---|---|
| Analysis subjects | 90,297 vitiligo patients + 90,297 controls |
| Diagnosis codes scanned | 1,265 codes (2007-2017) |
| Diseases with increased risk | 45 conditions (autoimmune diseases, melanoma, obsessive-compulsive disorder, etc.) |
| Diseases with decreased risk | 6 conditions (femoral fracture, etc.) |
Q3 Why does this screening approach matter?
The greatest significance is that it allows the data to reveal signals on its own, without being confined by assumptions. People naturally tend to look for associations that 'seem likely.' This method brings even connections no one suspected into consideration. In an era of large-scale data, this comprehensive scanning approach can serve as a starting point for new discoveries. Of course, automatically filtered associations may include chance or bias. The task remains that each one must be verified again through separate studies. For me, it was an interesting attempt to let the data tell me 'what to ask.'
Closing remarks
Without preselecting answers, we scanned the entire dataset. This allowed us to retrieve unexpected associations linked to vitiligo as candidates. These signals provide good clues for future in-depth research. As corresponding author, I find satisfaction in proposing a new exploration method suited to the era of large-scale data.
