A prospective single-center cohort study was conducted at the Gastroenterology outpatient clinic of the Second Affiliated Hospital of Xi’an Jiaotong University. Between April 2023 and September 2024, 425 patients with H. pylori infection, who were initiating their first course of eradication therapy, were consecutively enrolled.

In this study, LTFU was defined as failure to complete the recommended confirmatory test within 8 weeks after completing the initial eradication therapy, or adherence to less than 80% of the prescribed medication. This definition is adapted from established criteria in the field to ensure accurate assessment of initial treatment failure versus later reinfection (11, 18–21).

Occupation was classified into a binary variable: government employee VS. Non-government employee (which encompassed all other occupations, including self-employed, corporate workers, farmers, etc.). This was done to test the potential influence of socioeconomic and occupational factors on follow-up adherence.

Payment method was categorized as “Out-of-pocket” or “Medical insurance” to evaluate the potential influence of financial burden on follow-up adherence.

Time to hospital was defined as the patient’s self-reported one-way travel time from their residence to the hospital under normal traffic conditions, measured in minutes. This measure reflects the actual door-to-door duration, which may include in-transit time and potential waiting periods, rather than mere geographical distance.

This prospective cohort study was conducted in accordance with the principles of the Declaration of Helsinki and received approval from the Medical Ethics Committee of the Second Affiliated Hospital of Xi’an Jiaotong University (Approval No. 2023231). The study protocol was prospectively registered with the Chinese Clinical Trial Registry (Registration number: ChiCTR2500096567). We received informed consent from all participants.

All participants completed a structured questionnaire under the guidance of uniformly trained research staff. The questionnaire was developed based on the survey instrument used in the National Epidemiological Investigation of H. pylori Infection in China. Data were collected through one-on-one interviews, and questionnaires were reviewed on-site upon completion to ensure completeness. In cases of missing or unclear information, follow-up contacts were made to supplement the data; otherwise, incomplete questionnaires were excluded from the analysis.

The data were entered and verified using EpiData 3.2 software. All statistical analyses were performed via the R statistical package and EmpowerStats 4.2 software. Continuous variables were expressed as mean ± SDs and compared using the Student’s t-test or Mann–Whitney U test, as appropriate. Categorical variables were presented as frequencies and percentages and compared using the chi-square test or Fisher’s exact test.

The Least Absolute Shrinkage and Selection Operator (LASSO) regression method was applied to screen potential predictors. Variables with non-zero coefficients in the LASSO regression were subsequently entered into a multivariate logistic regression model to construct the final prediction model (22). The model’s discriminative ability was evaluated using the area under the receiver operating characteristic curve (AUC), with internal validation performed via the bootstrap method (500 resamples). Calibration curves and a decision curve analysis (DCA) were used to measure the predictive performance of the nomogram. Clinical utility was assessed using decision curve analysis (DCA). p < 0.05 was considered statistically significant.

The flow chart of patients throughout this analysis is reported in Figure 1. The demographic and clinical characteristics of the participants are presented in Table 1. Among the 425 patients initially enrolled, 41 were excluded for the following reasons: 9 patients provided invalid questionnaires, 1 patient was excluded due to pregnancy, and 31 patients could not be contacted (refused to respond or provided invalid phone numbers). The final analytical cohort therefore consisted of 384 participants, who were randomly allocated into a training group (n = 268) and a test group (n = 116). In the training and test groups, 40.67% (109/268) and 31.03% (36/116) of participants were LTFU, respectively. Common reasons for attrition included being too busy at work, failure to initiate medication, and resolution or absence of symptoms. After comparing the baseline data of both groups, it was found that there were no significant differences in sex, age, BMI, type of work, residence, payment method, occupation, education, marital status, monthly household income per capita, smoking status, alcohol consumption, distance to the hospital, outpatient waiting time, time to hospital, and various medical histories (prior H. pylori infection, family history of H. pylori, family history of gastric disease), indicating a generally comparable outcome distribution between the two sets for model building.

To minimize the risk of overfitting, a LASSO regression analysis was employed. This approach identified seven robust predictors of LUFU from the 19 baseline variables (Table 1; Figure 2): BMI, occupation, distance to the hospital, outpatient waiting time, alcohol consumption, type of residence, and awareness of the need for follow-up.

After multivariate analysis, the independently statistically significant predictors of LTFU were BMI, occupation, distance to hospital, outpatient waiting time, alcohol consumption, and awareness of follow-up (Table 2). The final model is expressed as: logit (LTFU) = −7.29047 + 1.84325*BMI + 1.17386*occupation +0.99756*alcohol consumption +2.87117*distance to hospital +0.01744*outpatient waiting time +1.49824*awareness of follow-up (Figure 3).

The prediction model is verified internally. Internal verification is done through bootstrapping, times = 500. In the training set, the model demonstrated an AUC of 0.885 (95% CI: 0.843–0.918), with a sensitivity of 93.58% and a specificity of 67.92%. Comparatively, in the test set, the model achieved an AUC of 0.862 (95% CI: 0.794–0.925), with a sensitivity of 83.33% and a specificity of 77.50% (Figure 4). The calibration curves showed that the predicted probabilities were closely aligned with the observed outcomes across a range of predicted probabilities, confirming the model’s accuracy in estimating LTFU risk (Figure 5). Effectively forecasting the risk of patient LTFU in H. pylori eradication.

To evaluate clinical utility, DCA was performed on both the training and validation cohorts. As shown in Figure 6, the DCA results demonstrated that the proposed model yielded favorable clinical utility in both the training and validation cohorts, providing meaningful support for individualized risk assessment and clinical decision-making across reasonable threshold ranges.

Consistent with prior evidence (23, 24), our study found a travel distance of >10 km to the healthcare facility as an independent risk factor for LTFU. Patients residing farther away face greater time commitments and financial costs for follow-up visits, which may further reduce adherence to post-eradication assessments. Furthermore, Denis Opio et al. (25) found that long outpatient waiting time was a risk factor for LTFU, which is consistent with our study.

Alcohol consumption is a known risk factor for H. pylori management (26, 27). In our study, alcohol use significantly increased the likelihood of LTFU, with drinkers exhibiting a 1.82-fold higher risk than non-drinkers. This is consistent with previous research (28, 29). One possible explanation is that alcohol use often co-occurs with poor mental health, including anxiety (30) and depression (31), which may diminish a patient’s motivation to attend follow-up appointments.

Obesity is reported to increase the incidence of LTFU, Torrens C. et al. (32) found severe obesity (BMI > 35 kg/m²) was associated with a 2.44-fold increase in LTFU risk. Vejrup K. et al. (33) also found that individuals with higher BMI were more likely to be LTFU. These findings are in line with our results. Obesity is closely linked to metabolic syndrome (34), and patients with obesity often carry a higher burden of comorbidities, which may further affect their ability or willingness to adhere to follow-up schedules. However, contrasting results have been reported by Soedarsono et al. (35) and Frijtersa et al. (36), who found that underweight BMI was associated with LTFU. Thus, the relationship between BMI and LTFU warrants further investigation.

Interestingly, we observed unexpectedly that civil servants were more likely to be lost to follow-up compared with other occupational groups. Similarly, consistent with Mulat A et al. (37), who also reported higher LTFU among government employees. This may reflect competing demands and institutional constraints rather than lack of access: despite having health coverage and formal leave, rigid work schedules and high occupational demands may impede attendance for non-acute appointments. Additionally, reliance on annual employer-organized health checks might lead to underestimation of the need for timely, condition-specific follow-up. In contrast, Soedarsono et al. (35) and Frijters et al. (36) found that unemployment or informal employment was associated with LTFU. Therefore, the influence of occupation on LTFU remains complex and merits further research.

The primary value of this nomogram lies in enabling risk-stratified patient management. By providing a quantitative, individualized risk score at the point of prescription, it shifts the clinical approach from a uniform, reactive follow-up protocol to a proactive, resource-efficient strategy. For patients stratified as high-risk, targeted interventions can be immediately deployed: (1) Enhanced Pre-treatment Counseling: Structured education emphasizing the critical importance and specific timing of the confirmatory test; (2) Proactive Engagement: Implementing scheduled check-in calls or messages during the treatment course to address side effects and reinforce adherence. (3) Reducing Logistical Barriers: Exploring flexible scheduling, telehealth options for post-therapy consultation, or facilitating testing at more convenient locations. (4) Digital Reinforcement: Automated SMS or application-based reminders tailored to high-risk individuals.

This approach optimizes the allocation of limited clinical resources (e.g., nursing time, digital health infrastructure) to the patients who need them most, aiming to improve follow-up completion rates and, consequently, long-term clinical outcomes.

In defining LTFU as a missed confirmatory test within 8 weeks post-therapy, we aimed to balance clinical guidelines with diagnostic accuracy. This operational definition, adapted from established criteria (11), is designed to maximize the likelihood that a positive test result within this window reflects true eradication failure rather than early reinfection, a distinction that becomes less clear beyond this period (21). While this timeframe is clinically reasoned for our setting, its optimality may vary in different healthcare contexts and remains a parameter for consideration in future research or implementation.

Key strengths of this study include its prospective design, the incorporation of diverse and clinically relevant predictors, rigorous internal validation using bootstrapping, and evaluation of clinical utility via decision curve analysis. The nomogram offers an intuitive, visual tool that could be integrated into routine practice. The study has several limitations. Its prospective cohort design is inherently resource-intensive. Moreover, the reliance on questionnaire-based data collection led to the exclusion of participants with incomplete responses, potentially introducing selection bias. Most importantly, the predictive nomogram was developed and internally validated within a single-center context. Consequently, its generalizability to other healthcare settings remains unestablished and warrants external validation through future multi-center studies.