Research on what disease screening can do

There is no need to worry about whether you should sift, what sieve to use, and what to do after sifting. It is valuable to focus on the pain points of landing.

The easiest and most practical thing to produce results now is to study the boundaries of suitable groups for screening. After all, the controversy about "over-screening" has never stopped. Take prostate cancer PSA screening as an example. Europe and the United States have stopped universal screening before, saying that many low-risk elderly people had their prostates removed after testing positive. Instead, they suffered the sequelae of urinary incontinence and sexual dysfunction, and there was no benefit at all. However, a domestic cohort study published by the Fudan Public Health Team last year showed that PSA screening for high-risk groups aged 55-69 with a family history of prostate cancer can reduce the mortality rate of late-stage prostate cancer by 32%. How to determine the threshold for the population in between? For diseases that are also subject to overdiagnosis, such as thyroid cancer and breast nodules, a localized cohort study can be conducted. It does not require a large sample. It can clearly characterize the local high-risk groups and provide reference standards for the disease control department, which is more practical than anything else.

If you don’t want to do too macro-scale population research, there is no problem in going in the direction of technology implementation. Nowadays, new technologies for early screening are coming out quickly, and methylation sequencing and AI image-assisted diagnosis all have beautiful data in the laboratory. However, if they are really used in grassroots screening scenarios, they will be "acclimated". For example, last year, Guangzhou Liwan did a colorectal cancer screening pilot. The sensitivity of the original fecal occult blood test was only 60%. It was replaced. The sensitivity of the new fecal methylation reagent has reached 91%, but the cost of a single reagent has increased by 120 yuan. The project team was stuck here at the time: If this reagent is included in routine local public health screening, it will cost nearly 80 million more per year, but it can detect more than 2,000 more cases of early cancer. How to calculate this account? This kind of health economics research rooted in real scenarios, both the medical insurance department and early screening companies are rushing to refer to it, which is much more valuable than adjusting parameters in the laboratory. Of course, some people say that now we should improve the technical sensitivity first and then talk about the cost. Both sides are reasonable. If we can improve the technical adaptability and reduce the cost at the same time, it is also a good research direction.

There is another type of research that was previously ignored but is now increasingly needed, which is full-link management after screening. Two years ago, I came into contact with a team from a community health service center in Quzhou, Zhejiang. They looked at the local fundus disease screening data for diabetic patients and found that only 28% of the patients who tested positive actually went to higher-level hospitals for further examinations. The rest either didn't take it seriously or didn't know which department to go to, so the screening was in vain. Later, they conducted a small pilot project, assigning dedicated family doctors to patients with early stage lesions, sending follow-up reminders every month, and opening up a green referral channel for ophthalmology departments in the city. In half a year, the follow-up rate increased to 87%. They also helped more than 30 patients with proliferative lesions undergo intervention, and almost went blind. This kind of small-scale intervention research can be directly replicated in other communities, which is very down-to-earth.

Nowadays, many people are rushing to build AI-assisted screening models. Most of them use public data to train a model, brush up the accuracy and then publish a paper. When it comes to community health service centers, grassroots doctors can’t even understand the parameters output by the model and dare not use it at all. If they can concentrate on optimizing AI screening tools for grassroots adaptation, it is also a blue ocean direction that no one is grabbing.