Understanding Serous Borderline Ovarian Tumour Recurrence: Risk Factors, Prediction Tools, and What It Means for Patients

Table of Contents

• Key Points
• What Is Serous Borderline Ovarian Tumour (SBOT)?
• Why This Research Matters
• Who Took Part in the Study
• How the Researchers Collected and Analysed the Data
• The Five Key Risk Factors for Recurrence
• Building a Recurrence Prediction Tool
• A Simpler, Seven-Variable Model for Everyday Use
• How the Random Forest Model Compared With Other Methods
• Molecular Findings: What Happens Inside Recurrent Tumours
• How the Laboratory Analyses Were Performed
• Clinical Implications: What This Means for Patients
• Study Limitations: What This Research Could Not Prove
• Recommendations and Next Steps for Patients
• Funding Sources
• Frequently Asked Questions
• Source Information

Key Points

• Serous borderline ovarian tumour has an excellent prognosis, with five-year survival above 96%.
• Five independent recurrence risk factors are micropapillary pattern, advanced stage, fertility-sparing surgery, microinvasion, and lymph node invasion.
• Fertility-sparing surgery is linked to recurrence rates of 25.0% to 56.4% in reported studies.
• A seven-variable online tool predicts recurrence risk and performed well in internal and external patient groups.
• Recurrent SBOT may suppress the immune system through metabolic changes, suggesting possible targets for future therapy.

What Is Serous Borderline Ovarian Tumour (SBOT)?

Borderline ovarian tumour (BOT) is a rare type of ovarian neoplasm. Unlike fully malignant ovarian cancers, BOT does not invade the underlying stromal tissue (the supportive framework of the ovary). Instead, it features atypical epithelial cell proliferation — meaning the cells lining the ovary surface grow abnormally, but not invasively.

Patients with borderline tumours tend to be much younger than those with malignant ovarian cancer, and their prognosis is considerably better. The overall 5-year survival rate for patients with BOT at any stage is higher than 96%.

Serous borderline ovarian tumour (SBOT) is the most common subtype, accounting for about 50% of all borderline ovarian tumour patients. The word "serous" refers to the type of cell involved, which normally produces a watery, serum-like fluid. Despite the generally favourable outlook, SBOT at stage II or higher can still progress to a more dangerous condition called low-grade serous carcinoma (LGSC). Because of this risk, surgical removal of the tumour is considered the first-line therapy.

This means that for most patients, surgery is the main treatment. The central question is how aggressive that surgery needs to be — and that depends largely on the risk of the tumour coming back (recurring) after treatment.

Why This Research Matters

A major dilemma faces women with SBOT who are of reproductive age and wish to have children in the future. Guidelines from the National Comprehensive Cancer Network (NCCN) and several clinical studies indicate that fertility-sparing surgery (FSS) is a valid option for these patients. FSS aims to preserve the uterus and at least part of one ovary, allowing the possibility of future pregnancy.

The catch is that choosing FSS increases the risk of tumour recurrence. In published literature, recurrence occurs in 25.0% to 56.4% of patients with SBOT who undergo this approach. If doctors could accurately estimate a patient's recurrence risk before surgery, they could more carefully match each woman to the best type of surgery — and design a personalised follow-up plan.

Unfortunately, most clinical studies on SBOT recurrence conducted before this one had important weaknesses. They often involved relatively small patient groups followed for only short periods. The largest SBOT study worldwide used clinical data from 1,026 Danish patients. For the Chinese population, the largest previous study included only 101 patients, of whom 12 (11.9%) developed tumour recurrence over a follow-up interval of 7 years.

Studies on the molecular features of SBOT were even more limited. Most focused on genomic mutations found in small numbers of patients and relied on laboratory assays with limited resolution and throughput. There was a clear lack of systematic research connecting SBOT's clinical features to its genomics (the full DNA blueprint of the tumour), transcriptomics (which genes are switched on), metabolites (small molecules produced by cellular processes), and lipids (fats and fat-like molecules).

Identifying molecular features associated with recurrence matters because it may reveal mechanisms that can be targeted by new drugs. The researchers behind this study hypothesised that a large patient cohort, followed for a long period and analysed with advanced computational and experimental methods, could uncover important features of recurrent SBOT.

Who Took Part in the Study

The study used two separate groups of patients, called cohorts.

The internal cohort (the main group used for discovery and model training) included 321 patients who underwent surgery for primary or post-relapse serous borderline ovarian tumour between August 1, 2009, and July 31, 2019, at the Obstetrics and Gynecology Hospital of Fudan University in Shanghai, China. Two of these patients were excluded because they were diagnosed with SBOT alongside invasive implantation (a condition that, under the 2020 World Health Organization Classification of Tumors of Female Reproductive Organs, should be treated as low-grade serous carcinoma). This left 319 patients in the internal cohort.

The external cohort (used to test whether the prediction model works on independent data) consisted of 100 patients diagnosed with SBOT after surgery at the Shandong Provincial Hospital in China, collected between August 1, 2009, and November 30, 2019.

Together, the patients came from 23 of China's 31 provinces, which makes the findings broadly representative of the Han Chinese population.

The study was approved by the Ethics Committee of the Obstetrics and Gynecology Hospital of Fudan University (approval number OBGYN 2019-70). All participants gave written informed consent — meaning they were fully told what the research involved and agreed to take part.

How the Researchers Collected and Analysed the Data

Two researchers collected basic patient information, including medical files and follow-up data. Every tumour's pathology results were independently re-confirmed by two pathologists who did not know the original diagnosis — a "blind" review designed to reduce bias. Tumour staging followed the 2014 International Federation of Gynecology and Obstetrics (FIGO 2014) criteria.

Treatment decisions were guided by patient age and fertility wishes. Patients younger than 40 years old, or who had never given birth (nulliparous), were offered fertility preservation for either their initial treatment or their recurrence surgery. Fertility-sparing surgery (FSS) was used to keep the uterus and at least one whole ovary. It took several forms:

• Unilateral ovarian cystectomy (UOC) — removing only the cyst from one ovary
• Unilateral salpingo-oophorectomy (USO) — removing one ovary and one fallopian tube
• Bilateral ovarian cystectomy (BOC) — removing cysts from both ovaries
• Unilateral salpingo-oophorectomy plus contralateral ovarian wedge resection (USO + CWR) — removing one ovary and tube plus a wedge-shaped piece of the other ovary

For patients without a desire to preserve fertility, the standard approach was total hysterectomy with bilateral salpingo-oophorectomy (TH + BSO) — removal of the uterus, both ovaries, and both fallopian tubes. Additional surgical steps were chosen selectively, including removal of pelvic lesions, multiple biopsies of the peritoneum (the membrane lining the abdominal cavity), omentectomy (removal of the omentum, a fatty abdominal membrane), and lymphadenectomy (removal of lymph nodes).

If the SBOT was on one side only, tumour size was recorded as the maximum diameter. In bilateral cases, the size was defined as the sum of both tumours' diameters. Following NCCN guidelines, "standard cytoreductive surgery" meant removal of all visible (macroscopic) lesions, peritoneal biopsy, peritoneal washings, and omentectomy. Surgery was classed as "incomplete" if any of these standard steps was missed.

Patients attended follow-up appointments according to a fixed schedule: every 3 months during the first year after surgery, every 6 months during the second year, and every 12 months after that. At each visit, doctors recorded recurrence status, physical examination findings, tumour markers, ultrasound imaging, and magnetic resonance imaging (MRI) results.

Recurrence was strictly defined as either a serous borderline lesion or low-grade serous carcinoma confirmed by repeat surgery, or a highly suspicious pelvic mass seen on MRI together with an elevated level of the tumour marker CA125 (carbohydrate antigen 125). Progression-free survival time was the interval between the primary operation and the date recurrence was discovered on imaging.

The Five Key Risk Factors for Recurrence

This is the heart of the study's clinical findings. Using survival analysis, the researchers first tested each clinical factor individually with univariate Cox analysis (a statistical method that examines how a single variable relates to the time until an event like recurrence). Factors that showed a significant link were then entered into a multivariate Cox analysis, which accounts for multiple variables at once to find out which ones are independently important.

One detail is worth noting: because there was only a single stage 4 patient in the cohort, and that patient did not experience recurrence (which could distort the hazard ratio — the measure of risk — for the "stage" indicator), this one patient was removed before the Cox analysis.

The analysis identified five factors significantly correlated with SBOT recurrence in this Han Chinese population:

1. Micropapillary pattern — a specific architectural pattern seen under the microscope in which the tumour cells form very small, finger-like projections
2. Advanced stage — the tumour has spread beyond the ovary at the time of diagnosis (stage II or higher under FIGO 2014 criteria)
3. Fertility-sparing surgery (FSS) — surgery that leaves part of the reproductive organs intact rather than removing them completely
4. Microinvasion — the presence of tiny, microscopic areas where tumour cells have begun to penetrate the underlying tissue, visible only under a microscope
5. Lymph node invasion — tumour cells found in the lymph nodes

These results were displayed as survival curves showing how each significant feature affected recurrence rates over time.

For patients, the practical message is clear: if a pathology report mentions microinvasion or lymph node involvement, or if the tumour shows a micropapillary pattern, the recurrence risk is higher, and closer follow-up may be warranted. FSS carries a trade-off that each patient should discuss carefully with her surgical team.

Building a Recurrence Prediction Tool

The researchers wanted to move beyond simply listing risk factors. Their goal was to build a practical tool that could predict, for an individual patient, the probability of recurrence.

The internal cohort of 319 patients contained 28 clinical indicators per patient. These included demographic characteristics (age, region), fertility status, clinical information, laboratory test results, surgical details, and pathological information. After removing 48 samples that lacked a clear recurrence or non-recurrence label, 271 samples remained, including 59 patients who recurred (about 1 in 5).

Because only 59 of 271 patients (roughly 22%) experienced recurrence, the data were class-imbalanced — meaning there were far more non-recurrence cases than recurrence cases. Left unaddressed, this would bias the model toward predicting "no recurrence" for everyone. To correct this, the researchers used a technique called random oversampling, in which the 59 recurrent samples were duplicated randomly to expand the recurrence group. Sensitivity analysis confirmed this approach was sound. The final training dataset contained 424 samples: 212 recurrence cases and 212 non-recurrence cases.

Two-thirds of the internal cohort was randomly selected as the training set to construct a Random Forest Regressor — a machine learning algorithm that builds many decision trees and averages their outputs. The remaining one-third served as the internal test set. During training, the model was tested more than 20 times to make sure its predictions were stable.

For the external validation, the same label-filtering process was applied to the 100 external samples, leaving 83 patients — including 24 who recurred (about 29%). Missing values in the clinical data were handled using a technique called "missForest" imputation, which estimates missing values based on patterns in the other data. The three indicators with the most missing data were HE4 (a tumour marker, missing for 39.83% of all missing values), CA199 (another tumour marker, 12.33%), and aspartate aminotransferase or AST (a liver enzyme, 8.3%). Sensitivity analyses, in which 1–8% of data were artificially removed at random, confirmed that the imputation was reliable.

Because the model produced a continuous score between 0 and 1, a threshold of 0.5 was chosen. Patients with a predicted value above 0.5 were classified as high risk for recurrence; those at or below 0.5 were classified as low risk. The model performed strongly:

• Internal cohort: Area Under the Curve (AUC) of 0.999 — essentially near-perfect discrimination between patients who would and would not recur
• External independent cohort: AUC of 0.817 — still highly accurate when applied to a completely separate group of patients

(For readers unfamiliar with statistics: the AUC measures how well a test distinguishes two groups. An AUC of 0.5 would be no better than a coin flip; 0.999 and 0.817 indicate excellent and strong predictive accuracy, respectively.)

A Simpler, Seven-Variable Model for Everyday Use

The researchers recognised a practical problem: collecting all 28 clinical indicators for every patient is burdensome in busy medical settings. They therefore quantified the importance of each variable using a metric called the mean decrease Gini score (a measure of how much each variable contributes to the model's accuracy). Variables with higher scores play more important roles in prediction.

When the ranked scores were plotted as a curve, the improvement flattened out after the seventh variable. This suggested that just seven key indicators could provide nearly all the predictive power. The team screened out those seven variables and trained a simplified model.

Both versions — the full 28-indicator predictor and the simplified one — are available free online at http://117.25.169.110:1030/. The website was built using the Gin web framework with Bootstrap-designed pages, making it accessible to medical workers. A doctor can enter a patient's clinical information and receive an instant estimate of recurrence probability.

How the Random Forest Model Compared With Other Methods

The research team wanted to confirm that their choice of algorithm was the best one for this data set. They tested four popular supervised machine-learning methods:

• lightGBM — a fast gradient-boosting framework: achieved AUC of 0.950
• XGboost — another gradient-boosting method: achieved AUC of 0.971
• gcForest — a deep forest algorithm: achieved AUC of 0.986
• Random Forest Regressor — the method ultimately chosen: achieved AUC of 0.999

Each model was fed the training data and tuned over 20 rounds to maximise its AUC. The Random Forest Regressor showed the best stability and highest accuracy with this dataset, which is why it was selected for the final online tool.

Molecular Findings: What Happens Inside Recurrent Tumours

Beyond clinical factors, the researchers wanted to understand the biological machinery driving recurrence. They performed a "multi-omics" analysis on the original SBOT samples — that is, they examined the tumours at several molecular levels simultaneously:

• Whole genome sequencing (WGS) — reading the complete DNA sequence of the tumour to spot mutations
• RNA-seq (transcriptomics) — measuring which genes are actively expressed in the tumour tissue
• Metabolomics — profiling the small molecules (metabolites) produced by the tumour's metabolism
• Lipidomics — profiling the fats and fat-like molecules in the tissue

The central finding was that recurrence of SBOT is related to metabolic regulation of immunological suppression. In plain language: tumours that later recur appear to reprogram their metabolism in ways that suppress the immune system's ability to attack them. The tumour essentially creates a protective microenvironment by altering its metabolic "fuel" usage, which switches off local immune responses.

This is a clinically valuable insight because metabolic pathways and immune checkpoints are both targetable with drugs. If a recurrent SBOT suppresses immunity through specific metabolic changes, medications that reverse those changes might, in the future, reduce recurrence risk or help treat recurrences when they happen.

Quantitative RT-PCR (a laboratory technique that measures gene expression levels in real time) was used to confirm the expression of key genes, comparing target gene expression against the reference gene 36B4 for standardisation. Immunohistochemical analysis was also performed: tumour sections were examined under a 10X low-power lens to locate the tissue, then ten randomly chosen fields at 40X high power were used to count immune cells in each sample. The average cell counts were compared between recurrent and non-recurrent groups.

How the Laboratory Analyses Were Performed

The laboratory methods were rigorous and worth describing in brief, as they determine how much trust can be placed in the findings.

For whole genome sequencing: Approximately 50–100 mg of tissue was cut into pieces and digested in a lysis buffer containing Proteinase K (an enzyme that breaks down proteins) at 56 °C for 60–120 minutes. DNA was extracted using phenol/chloroform/isoamyl alcohol, precipitated with isopropanol at −20 °C, washed with 75% ethanol, and re-dissolved in TE buffer. DNA concentration was measured with a Qubit Fluorometer, and integrity was checked by agarose gel electrophoresis. One microgram of genomic DNA was randomly fragmented using Covaris equipment, and fragments of 200–400 base pairs were selected. After end-repair, adenylation, adapter ligation, and PCR amplification, the DNA was circularised and sequenced on a BGISEQ-500 platform (BGI-Shenzhen, China).

Raw sequencing data were cleaned by removing adapter sequences, reads with more than 50% low-quality bases, and reads with more than 10% unknown bases. Clean data were mapped to the human reference genome (GRCh38) using BWA version 0.7.17-r1188 in MEM mode. SNP calling used GATK v4.1.7.0, and variants were annotated with Annovar version 2019-10-24. The researchers used the Fst statistic (a population-genetics measure of genetic differentiation) — via Vcftools version 0.1.16 — to find variants specific to the recurrence group. When a variant is shared by all individuals in one group and absent from all individuals in the other, the Fst value equals 1, indicating a completely group-specific variant.

For RNA-seq: Total RNA was extracted using TRIzol reagent (Invitrogen). Libraries were generated with the NEBNext Ultra RNA Library Prep Kit for Illumina, with index codes added to each sample. Messenger RNA was purified using poly-T oligo-attached magnetic beads, then fragmented using divalent cations at elevated temperature. Complementary DNA (cDNA) was synthesised in two strands, purified with the AMPure XP system, and subjected to PCR amplification. Sequencing was performed on an Illumina HiSeq platform, generating 125–150 base-pair paired-end reads. Raw data were cleaned with Trimmomatic, aligned to the reference genome with Hisat2, and gene expression was quantified with StringTie. Differential expression analysis between the recurrent and non-recurrent groups was performed with the DESeq2 R package.

For quantitative RT-PCR: Total RNA (500 ng) was converted to cDNA using the PrimeScript RT reagent Kit (Takara), and PCR was run on an Applied Biosystems QuantStudio 5 instrument with SYBR Green PCR Master Mix. Fold changes were calculated relative to the reference gene 36B4.

For metabolomics: Tissue samples were homogenised at −20 °C for 1.5 hours. Pre-chilled 80% methanol in water was added, and after centrifugation, the supernatant was dried with a SpeedVac concentrator and stored at −80 °C. The dried samples were reconstituted in acetonitrile:water (50:50) and injected into a liquid chromatography-mass spectrometer (LC-MS). The targeted metabolomics method used an amide HILIC column with a specific mobile-phase gradient lasting 17 minutes.

For lipidomics: Tissue samples were homogenised with water and methanol, then mixed with methyl tert-butyl ether (MTBE) for lipid extraction. After vortexing, rocking for one hour at room temperature, and centrifugation, the top organic phase was collected and dried under a stream of nitrogen. Extracted lipids were stored at −80 °C before mass spectrometry analysis.

Clinical Implications: What This Means for Patients

For a woman diagnosed with SBOT, this study offers several practical messages. First, it confirms that not all SBOTs behave the same way. A patient whose pathology report shows micropapillary features, microinvasion, or lymph node involvement carries a meaningfully higher recurrence risk and may require more aggressive treatment or more intensive surveillance.

Second, the trade-off around fertility-sparing surgery is now clearer and more quantifiable. FSS raises recurrence risk, but it does so in a way that can be estimated before surgery using the online prediction tool. A young woman with a low predicted recurrence score might reasonably choose FSS, preserving her chance of biological children. A woman with a high predicted score might decide — with her doctors — that complete removal of the reproductive organs (TH + BSO) is the safer path, or that she needs particularly rigorous follow-up if she still wishes to pursue FSS.

Third, the molecular discovery that recurrence is tied to metabolic regulation of immunological suppression opens the door to future therapies. If scientists can identify the specific metabolic changes that help recurrent tumours hide from the immune system, drugs that target those pathways could potentially be developed. These findings will contribute to the development of personalised and targeted therapies to improve prognosis.

Fourth, the validation of the prediction tool on an independent external cohort (AUC 0.817) suggests the model is not overfitted to one hospital's data. It generalises to other institutions and could, after further validation, become a standard part of SBOT clinical care.

Study Limitations: What This Research Could Not Prove

Like all medical research, this study has limits that patients and clinicians should keep in mind. The study is retrospective, meaning it looked backward at medical records rather than following patients forward in a randomised trial. Retrospective designs can identify associations but cannot prove causation as definitively as prospective trials can.

The cohorts were drawn exclusively from Chinese hospitals and the findings apply specifically to a Han Chinese population. While patients came from 23 of 31 provinces, the results may not fully generalise to other ethnic groups or healthcare systems, though the researchers' use of an external validation cohort at a different hospital strengthens confidence in the model.

The molecular analyses (genomics, transcriptomics, metabolomics, lipidomics) were performed on original tumour samples. This means they reveal features present at the time of initial surgery, not at the moment of recurrence. The study links these baseline molecular features to later recurrence, but cannot describe what changes occur in the tumour between the first surgery and the recurrence itself.

The machine learning model, despite its impressive internal AUC of 0.999, was built on a relatively small number of recurrence events (59 in the internal cohort, 24 in the external cohort). The simplified seven-variable model, while more practical, may sacrifice some accuracy in patient subgroups not well represented in the training data. The website tool should therefore be viewed as a decision aid to support — not replace — clinical judgement.

Finally, the predicted molecular mechanisms (metabolic regulation of immunological suppression) are based on association studies. Experimental work will be needed to confirm exactly which metabolic pathways drive immune suppression and whether they can be safely targeted with drugs.

Recommendations and Next Steps for Patients

For patients diagnosed with SBOT, the following practical steps emerge from this research:

1. Ask for a detailed pathology report. Specifically ask whether micropapillary pattern, microinvasion, or lymph node invasion were present. These three features, plus tumour stage, are the most important pathological determinants of recurrence risk.
2. Discuss fertility-sparing surgery honestly. If you are under 40 or have not yet had children, FSS is a legitimate option — but it carries a higher recurrence risk. Bring the statistics from this study (recurrence rate in the literature ranges from 25.0% to 56.4% with FSS) to your conversation with your surgical team.
3. Consider having your recurrence risk estimated with the online tool (http://117.25.169.110:1030/). Ask your doctor whether the seven-variable model is appropriate for your case. Use the result as one input among several in your treatment decisions.
4. Follow your surveillance schedule. The study used follow-up every 3 months in year one, every 6 months in year two, and annually thereafter, with CA125 blood tests, ultrasound, and MRI. Even if your predicted risk is low, adhere to this schedule — recurrence seen early is generally easier to treat.
5. Join a clinical registry if possible. Larger cohorts and longer follow-up will make the prediction tool even more accurate. If your hospital offers research participation, consider enrolling.
6. Stay informed about targeted therapies. The finding that recurrent SBOT involves metabolic regulation of immune suppression means future treatment may include metabolic modulators or immunotherapy combinations. Ask your oncologist whether any relevant trials are open at your centre.

Funding Sources

This research was supported by multiple grants. Jin Li received funding from the Chinese Ministry of Science and Technology (MOST) under grants 2020YFA0803600 and 2018YFA0801300, the National Natural Science Foundation of China (NSFC) under grant 32071138, and SKLGE-2118. Junqiu Yue was funded by the Initial Project for Young and Middle-aged Medical Talents of Wuhan City, Hubei Province ([2014] 41). He Huang received funding from MOST grants 2019YFA0801900 and 2020YF1402600. Jia Song was funded by NSFC grant 22104080. Chao Gu received funding from the Natural Science Foundation of Shanghai (20ZR1408800) and NSFC grant 82171633. Bin Li was funded by the Natural Science Foundation of Shanghai (19ZR1406800).

Frequently Asked Questions

What is the survival rate for serous borderline ovarian tumour?

For patients with borderline ovarian tumour at any stage, the overall five-year survival rate is higher than 96%. This includes all types, and SBOT is the most common subtype. Although the outlook is excellent, recurrence can still happen, especially in certain higher-risk situations.

What are the main risk factors for SBOT recurrence?

In a Chinese study of 319 patients, five independent risk factors were identified: a micropapillary pattern, advanced stage, fertility-sparing surgery, microinvasion, and lymph node invasion. If your pathology report mentions any of these, your care team may recommend closer follow-up or more extensive surgery.

Does fertility-sparing surgery for SBOT increase recurrence risk?

Yes. Fertility-sparing surgery leaves part of the reproductive organs intact. In published literature, recurrence occurs in 25.0% to 56.4% of patients with SBOT who choose this approach. Discuss the trade-off carefully with your doctors, especially if you wish to have children in the future.

Is there an online tool to estimate my recurrence risk?

Researchers developed a free online prediction tool based on 28 clinical indicators, with a simplified version using only seven. It was accurate in both internal and external patient groups, with an AUC of 0.999 and 0.817, respectively. Ask your doctor if this tool could help guide your treatment plan.

What did molecular analysis reveal about recurrent SBOT?

Analysis of original tumour samples found that recurrent SBOT is associated with metabolic regulation of immunological suppression. In other words, tumours that later recur may change their metabolism to suppress the immune system's ability to attack them. This offers clues for future personalised treatments.

When should I seek a second opinion before choosing fertility-sparing surgery for serous borderline ovarian tumour?

Seeking a second opinion is advisable before deciding between fertility-sparing surgery and complete removal of reproductive organs, because the choice directly changes recurrence risk. In serous borderline ovarian tumour, fertility-sparing surgery has a reported recurrence rate of 25.0% to 56.4%. Your pathology should be reviewed for micropapillary pattern, microinvasion, lymph node invasion, and FIGO stage, since these risk factors influence prognosis. An online seven-variable model can estimate individual recurrence risk. Independent second opinion services can review pathology and surgical recommendations. Diagnostic Detectives Network provides independent expert second opinions.

Source Information

Original article title: Identification of clinical and molecular features of recurrent serous

Authors: Ziyang Lu, Fanghe Lin, Tao Li, Jinhui Wang, Cenxi Liu, Guangxing Lu, Bin Li, MingPei Pan, Shaohua Fan, Junqiu Yue, He Huang, Jia Song, Chao Gu, and Jin Li. (Fanghe Lin, Tao Li, and Ziyang Lu contributed equally to this work.)

Journal: eClinicalMedicine (The Lancet Discovery Science), 2022; Volume 46, Article 101377. Published online April 8, 2022.

DOI: https://doi.org/10.1016/j.eclinm.2022.101377

Corresponding authors: Jia Song (songjiajia2010@shsmu.edu.cn), Chao Gu (chaogu@fudan.edu.cn), and Jin Li (li_jin_lifescience@fudan.edu.cn).

Copyright: © 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

This patient-friendly article is based on peer-reviewed research. It is intended for educational purposes and does not constitute medical advice. Patients should discuss all treatment decisions with their own healthcare team.