This miracle doctor just wants to be fired!

In the second week after the three students joined the group, Zhou Cheng asked them to submit a research proposal each.

The requirements are simple; there are no restrictions on the direction or the requirement to rely on existing research topics.

The only condition is that the research question must come from clinical practice; you can't just find a gap in the literature and then try to fill it in.

Friday afternoon.

The three people carried their laptops into the small conference room.

Lin Feng had just come down from the catheterization lab when Zhou Cheng called him over.

He originally thought it was just a regular group meeting, but when he entered, he found that the whiteboard had been wiped clean, there were three printed plans on the table, and Zhou Cheng had just brewed some tea next to it.

"Is everything ready?" Zhou Cheng asked.

All three nodded at the same time.

"Then let's proceed according to the order of joining the group." Zhou Cheng flipped to the top document. "Mingyuan, you go first."

Li Mingyuan was clearly the most well-prepared.

He connected the computer to the projector, and the first page showed the complete research framework, with a very formal title—Animal Experimental Verification and Long-Term Vascular Response Study of Calcification Ring Targeted Release Technique.

This technique is an operational approach that Zhou Cheng previously proposed for severe coronary artery calcification lesions.

By identifying the location of stress concentration in calcified rings through intravascular imaging, targeted treatment can be carried out to minimize ineffective dilation and damage to normal blood vessel walls.

Preliminary clinical results have been published, and several articles have attracted considerable attention, but the main evidence currently accumulated still comes from clinical cases and in vitro models.

Li Mingyuan wants to extend this technology to animal experiments.

"My plan is to establish an animal model of coronary artery calcification, comparing it with three treatment methods: traditional high-pressure balloon endovascular treatment, rotational atherectomy, and targeted release, to observe immediate lumen acquisition, vessel wall damage, and long-term intimal hyperplasia." Li Mingyuan stood in front of the projection screen, speaking slowly and clearly about each part. "After the operation, the animals will be sacrificed in batches at one month, three months, six months, and twelve months to complete histopathological examinations, and at the same time, OCT and IVUS will be used to assess the vascular healing."

He switched to another page and continued to introduce the methods for constructing animal models, sample size estimation, and statistical procedures.

To improve the stability of the model, he plans to first induce calcification through a high-fat diet combined with local vascular injury, and then conduct imaging screening to include only animals that form continuous ring-shaped calcification in the experiment.

The control group also had standardized equipment specifications, dilation pressure, and operation time, and even the postoperative antiplatelet regimen was included.

Lin Feng nodded halfway through the conversation, "This plan is quite comprehensive."

Zhao Sisi also admitted: "If it can be done, the article should be of a very high level."

After Li Mingyuan finished speaking, he sat down, his eyes filled with anticipation.

He researched a lot of information for this proposal, comparing six different animal calcification models. The final design he selected was almost identical to the official tender document.

Zhou Cheng didn't comment on the merits of the plan, but only asked one question: "How long do you expect it to take?"

Li Mingyuan had clearly considered this: "The initial model establishment takes six to nine months, the formal experiment takes about a year, and the final pathology and data analysis takes about two and a half years."

"Assuming the model succeeds on the first try?"

The meeting room fell silent for a moment.

Li Mingyuan hesitated for a moment, then nodded: "Yes."

Zhou Cheng continued, "What if the first batch of animals didn't develop stable ring-shaped calcifications?"

"Adjust the inducement plan and do another batch."

"How long will it take?"

"At least another six months."

What about animal mortality rates?

Li Mingyuan flipped to the preliminary experimental data at the back of the protocol: "The mortality rate in similar studies is about 15% to 30%. We are dealing with local coronary artery injury, so the perioperative risk may be even higher."

Zhou Cheng tapped the timeline on the screen: "How many years did your doctorate take in total?"

Three years.

This time, without Zhou Cheng asking further, Li Mingyuan realized the problem.

The research approach is comprehensive and the logic is rigorous enough, but it assumes everything is based on ideal conditions.

The animal model was successful, the calcification morphology met the requirements, the surgical procedure did not experience numerous failures, and the histopathology was completed on time.

However, if any link in the chain is delayed, the three-year period will be quickly exhausted.

What's more troublesome is that these models have difficulty reproducing real human coronary artery calcification stably.

Even if we eventually get the results, it may take a long time to explain the differences between the model and clinical practice.

"Let's put the proposal aside for now, we'll discuss it later." Zhou Cheng said.

Li Mingyuan couldn't tell if he was satisfied, so he could only close the computer first.

……

The second person to give a report was Zhao Sisi.

Her plan continues in her area of ​​expertise: to develop a risk prediction model for coronary intervention.

Compared to the relatively rough design I presented during the interview, this one is significantly improved.

She plans to collect patients' preoperative clinical data, laboratory indicators, coronary imaging features, interventional procedure information, and postoperative follow-up data, and use machine learning to predict major adverse cardiovascular events within one year.

The model not only includes traditional structured data, but also plans to incorporate radiomics features extracted from imaging videos.

"Currently, many risk assessments only use traditional indicators such as age, kidney function, and left ventricular ejection fraction, which do not make sufficient use of lesion morphology and procedural details," Zhao Sisi said. "Theoretically, combining intraoperative imaging and clinical data can improve predictive capabilities."

Her projection displayed the complete data processing flow.

From missing value imputation and feature selection to the division of training and validation sets, and the performance comparison between different algorithms, the charts are beautifully presented.

The study also included a model interpretation module to help doctors understand which variables have the greatest impact on the prediction results.

Although Chen Yu didn't understand algorithms very well, he thought the model structure was impressive.

Lin Feng flipped through the paper plan: "How many cases are you planning to admit?"

"We'll start with our hospital's interventional data database from the past five years, which is expected to include 12,000 cases. If the data quality is satisfactory, we'll then contact two partner hospitals for external validation."

"More than 10,000 cases is enough," Lin Feng said.

Zhao Sisi nodded: "According to the preliminary estimate, the number of major endpoint events can also meet the modeling needs."

Looking at the flowchart on the projector, Zhou Cheng asked, "Are the data from the three hospitals the same?"

"Most of the variables are the same."

"I'm asking about the data distribution."

Zhao Sisi paused for a moment.

Zhou Cheng moved the mouse over one of the sets of features:

"Our hospital has a high proportion of patients with complex coronary arteries and a high utilization rate of intracoronary imaging. Many patients are transferred here after their conditions were difficult to treat at other hospitals."

"The second hospital you contacted mainly handles routine elective interventions, while the third hospital handles a large number of emergency PCI procedures."

"The three hospitals have different patient demographics, disease complexity, and treatment practices. The model performed very well in our hospital, but what would happen if it were applied to the other two hospitals?"

"This can be determined through external verification."

"What if the verification results are poor?"

Zhao Si thought for a moment: "Recalibrate the model."

"If the calibration is done, do I need to redo it at another hospital?"

Zhao Sisi did not answer immediately.

Zhou Cheng frowned and continued, "Different hospitals have different testing practices. Some hospitals routinely perform IVUS, while others rarely use it. A certain variable might have a 10% missing rate here, but it could be 70% missing at another hospital. How do you plan to handle that?"

The model on the projection was still beautiful, but problems began to emerge from the details.

Zhao Sisi originally viewed multicenter data as a way to expand the sample size, but she did not seriously consider the differences in patient composition and treatment processes among different hospitals.

Algorithms can put data into the same table, but clinical practice will not become consistent just because the tables are standardized!

"I'll go back and reanalyze the data distribution across the centers," Zhao Sisi said.

Zhou Cheng didn't press further and let Chen Yu begin.

……