Development and external validation of a pragmatic nomogram for point-of-care urolithiasis screening in resource-limited settings of Laos: a multicenter retrospective cohort study Authors: Banchao Shen, Shuhan Yang, Peng Wu, Yifan Pan, Dayuan Huang, Xinjie Chen, Pan Philavong, Kham Sing Thonkhamtha, Lamngeun Singphandy, Ping Meng, Chunlin Hao, Weijun Qin, Lijun Yang, Jingliang Zhang, Fuli Wang Published: 2026 ABSTRACT BackgroundThe rising burden of urolithiasis in Southeast Asia and the lack of efficient risk-screening tools for low-resource primary care, such as those in Laos, motivated the development and external validation of a pragmatic nomogram integrating accessible environmental, behavioral, and clinical factors for in-hospital urolithiasis risk stratification, with long-term goal of application in grassroots screening.MethodsA retrospective study was conducted with 665 consecutive inpatients (≥18 years, local residence ≥6 months) at a Lao Military Hospital (March 2024–February 2025). Data were collected using standardized questionnaires and electronic medical records. Least Absolute Shrinkage and Selection Operator (LASSO) regression selected predictors; nomogram performance was assessed by discrimination (area under the curve (AUC)), calibration (Brier score and Hosmer–Lemeshow test), and clinical utility (decision curve analysis (DCA)).ResultsUrolithiasis prevalence was 23.0% (153/665). Independent predictors included age (30–59 years, OR = 1.54; ≥60 years, OR = 1.13), sex (OR = 0.59), rural residence (OR = 0.52), high water intake (>1,500 mL/day, OR = 0.02), high sodium consumption (>10 g/day, OR = 21.90), and elevated blood calcium (OR = 9.12). The nomogram presented stable predictive performance, with an AUC of 0.882–0.907, Brier score of 0.086–0.111 (acceptable calibration, Hosmer–Lemeshow p = 0.198–0.892), and optimal clinical threshold at 0.20.ConclusionsThis externally validated nomogram uses six accessible environmental, behavioral, and clinical factors to stratify urolithiasis risk in Laos. At present, it helps guide targeted referral and optimize s resource allocation for hospitalized patients in resource—limited settings, while its application in primary care and among broader Southeast Asian populations remains to be prospectively evaluated. Introduction Urolithiasis is a common clinical condition with a heterogeneous global distribution, affecting approximately 1%–20% of the global population. It has higher prevalence rates in North America (7%–13%) and Europe (5%–9%), with an overall prevalence of 1%–19.1% across Asia, varying greatly by geographic area and reflecting regional differences in climate, diet, and healthcare access (Aizezi et al., 2022; Liu et al., 2018). Within Asia, prevalence varies substantially, from 5%–19.1% in regions such as West, Southeast, and South Asia and developed East Asian countries like Japan and South Korea, to 1%–8% in other areas (Liu et al., 2018). This increasing trend is linked to population aging and wide adoption of diagnostic imaging, particularly computed tomography (CT) and ultrasonography (Skolarikos et al., 2025). In Southeast Asia, hot and humid weather leads to fluid volume depletion, increasing the risk of kidney stone formation and underscoring the need for region-specific public health interventions. However, epidemiological data are limited. Thailand reported an incidence of 16.9% in 1997 (Yanagawa et al., 1997), whereas Laos, despite having similar environmental conditions, lacks comprehensive studies on adults. The only available study was conducted in children, which found that 63% of stones were of renal and 37% bladder localization (Sayasone et al., 2004). Laos’ risk profile is compounded by its mountainous geography (80%), tropical monsoon climate (28 °C–35 °C) promoting urine concentration, and underdeveloped healthcare system (Wang et al., 2024). These factors, along with documented variations in stone composition and Asia’s growing urolithiasis burden (Lang et al., 2022), suggest substantial underdetection in Laos. Thus, systematic investigation of the local epidemiology and risk factors is necessary to develop effective prevention strategies. Urolithiasis is influenced by demographic, metabolic, and lifestyle factors (Akram et al., 2024). Untreated urolithiasis can lead to serious complications, such as renal impairment (Zhang et al., 2021). These risks are increased in resource-limited settings, such as Laos, which has limited rural healthcare access, inadequate diagnostic imaging, and low health literacy. Such challenges increase the risk of complications and exacerbate the health disparities. Accordingly, local risk factors must be identified and feasible screening tools must be developed in Laos. Clinical prediction models, such as nomograms, may improve screening efficiency; however, existing urolithiasis models are generally preoperative assessment tools that rely on laboratory test results—they not only lack generalizability across populations but also fail to provide dedicated models for rapid point-of-care risk assessment. Notably, primary healthcare facilities currently lack urolithiasis risk-assessment tools that can be completed solely with questionnaires and fingertip blood tests (Zhu et al., 2024; Okita et al., 2020). Against this backdrop, this study takes simple questionnaire-based scoring as the benchmark and compares it with advanced modeling strategies to provide replicable risk-prediction algorithms for residents in resource-limited areas to enable rapid risk stratification and targeted treatment. This study aimed to (1) analyze sociodemographic and laboratory factors associated with urolithiasis in Laos, such as age, sex, body mass index (BMI), residence, occupation, family history, hypertension, diabetes, daily water intake, sodium consumption, smoking, beer consumption, serum calcium, and uric acid levels, and (2) develop and validate a context-adapted nomogram as a foundational step toward future primary care deployment in Laos. This tool enables low-cost urological risk stratification via questionnaires and serum tests and has the potential to gradually guide efficient patient referrals after calibration. Materials & Methods Ethical approval This study is a retrospective secondary analysis of prospectively collected data during a prior hospital-based study on urolithiasis. This retrospective study was approved by the Ethics Committee of Lao People’s Army Hospital 103 (for the urolithiasis nomogram development and temporal validation cohort, Approval No. LRY202401042) and the Ethics Committee of Xijing Hospital, Air Force Medical University (for the external geographic validation cohort, Approval No. KY20253100-2). All participants signed informed consent upon information form collection, which explicitly indicated their agreement to the health-related research use of their survey data. Given the retrospective profile of the study and the use of de-identified data (to protect participant privacy), the respective ethics committees waived the need to obtain informed consent again. All study procedures adhered to the Declaration of Helsinki (1975) and its subsequent amendments. Study design and sample size This single-center retrospective study followed the TRIPOD guidelines. To guide the study design, historical unpublished electronic medical records (EMRs) from the Urology Department of Lao People’s Army Force Hospital 103 (2021–2023) were first analyzed, which indicated urolithiasis prevalence of 22.5%–27.8%. Thus, the expected prevalence was set at 25%. Six predefined predictors (age, sex, daily water intake, sodium consumption, serum calcium, and uric acid) were selected via literature review and guideline recommendations (Liu et al., 2018; Skolarikos et al., 2025). The minimum sample size was 240 (calculated via 10 events per variable (EPV) and 25% prevalence), consistent with prediction model standards (EPV ≥ 10) (Pavlou et al., 2016). Data collection and variable definitions Data were retrospectively extracted from the Urology Department inpatients (March–October 2024) via two sources: (1) previously collected standardized paper questionnaires (demographics, lifestyle, and laboratory biochemical indicators; completed by trained data collectors at bedside during the original prospective hospitalization) (Questionnaire S1) and (2) existing EMRs, including standardized admission histories and nursing assessment records. Variables included age, sex, BMI, urban/rural residence, physical labor, family history of stones, hypertension, diabetes, daily water intake, sodium consumption, smoking, drinking history, blood serum calcium (albumin-adjusted, ion-selective electrode method), and blood uric acid (uricase–peroxidase method). Imaging results (ultrasound, X-ray, and computed tomography (CT)) were independently evaluated by two radiologists (blinded to clinical data) per European Association of Urology (EAU) (2022) guidelines (Geraghty et al., 2023). Three data sets were obtained at different time points: the training cohort data (March–October 2024), the temporal validation data (same institution, later time period; November 2024–February 2025), and the external geographic validation data (independent Chinese cohort; April–June 2025). Key variables were defined as follows: BMI: weight (kg)/height (m)2 (World Health Organization (WHO), 2000). Age group: according to the epidemiological peak of urolithiasis (Skolarikos et al., 2024; Chinese Urological Association et al., 2024), participants were categorized into young (<30 years), middle-aged (30–59 years), and older (≥60 years) adults—consistent with the 30–60 years predominant age band in these references. Age was derived from official identification documents, precise to the year. Hypertension: systolic blood pressure ≥140 mmHg or diastolic ≥90 mmHg (Unger et al., 2020). Diabetes: fasting glucose ≥7.0 mmol/L or random glucose ≥11.1 mmol/L (American Diabetes Association, 2023). Water intake: <500, 500–1,500, or >1,500 mL/day (measured using standardized graduated cups) Sodium consumption: <5, 5–10, or >10 g/day, categorized according to WHO-recommended intake thresholds (World Health Organization, 2025). Sodium consumption denotes dietary salt intake, as captured by the 24-h recall questionnaire. Beer consumption: >500 mL/day for >6 months CT-confirmed stone: attenuation ≥300 HU, diameter ≥3 mm (Geraghty et al., 2023) Cases were defined as patients with newly diagnosed CT-confirmed urolithiasis who were first-time stone formers. Participants who had any prior urolithiasis or had undergone urological stone surgery were excluded. Controls were individuals free of nephrolithiasis on admission and throughout hospitalization, with no history of prior stone disease. Biochemical testing at both hospitals followed identical protocols: ion-selective electrode for serum calcium and uricase-peroxidase assay for uric acid, with uniform lot-calibrated calibrators and control sera for all samples. Data management and study population Paper questionnaire data were double-entered with cross-validation; outliers were identified using the interquartile range (IQR) method. For missing data, prespecified handling strategies were applied sequentially: participants with >20% missingness were excluded (Sterne et al., 2009); for the remaining cases, original records were reviewed for supplementation; residual missing values were imputed using simple imputation (for <5% missingness) or MICE (RStudio v3.17.0) for 5%–20% missingness (Assad et al., 2024). Eligible participants were ≥18 years old, were local residence ≥6 months, and had ≥80% complete data. The exclusion criteria were as follows: impaired decision-making, refusal to participate, questionnaire–EMR discrepancies, prior urological surgery/stone treatment, estimated glomerular filtration rate ≤60 mL/min/1.73 m2, Child–Pugh B/C liver disease, cardiovascular disease, and severe infection. Recurrent urinary tract infections (≥3 episodes/year, European Association of Urology, 2022) and congenital urinary tract anomalies (Geraghty et al., 2023) were also excluded, as these conditions represent secondary stone etiologies unrelated to primary metabolic nephrolithiasis (Skolarikos et al., 2025). Data collection was conducted in two phases: (1) an initial recruitment phrase for the training cohort and (2) a follow-up phase in which additional participants (1/3 of training set size) were recruited to form the temporal validation cohort, ensuring temporal separation from the training data. Additionally, an external geographic validation cohort (eligible participants from a Northwest China hospital with primary care constraints similar to those in Laos) was established to assess cross-regional generalization beyond Southeast Asia. Baseline characteristics of the three sets were compared to confirm that absence of significant differences among key variables. Statistical analysis Data were analyzed using IBM SPSS Statistics version 27.0 (IBM Corp., Armonk, NY, USA) and R 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria). Continuous variables underwent Shapiro–Wilk normality testing, reported as mean ± standard deviation (SD, normal) or median (IQR) (nonnormal). Categorical variables are presented as frequencies (%).Independent t-tests (normal continuous), Mann–Whitney U tests (nonnormal continuous), and chi-square tests (categorical) were performed for group comparisons. LASSO regression (glmnet 4.1-8) was used for predictor selection to address overfitting/collinearity, followed by multivariate logistic regression (rms v7.0-0) for nomogram development. Statistical significance was set at p < 0.05. Nomogram development and validation Multicollinearity was assessed via the generalized variance inflation factor (GVIF). Predictors with nonzero coefficients were selected via LASSO regression (K-fold cross-validation for stability). Variables with p < 0.1 in the univariate analysis underwent Akaike Information Criterion (AIC)-based backward stepwise regression to build the nomogram (Harrell Jr, 2026). The performance of the proposed model was subjected to internal validation, with bootstrapping (1,000 samples) used to calculate the area under the curve (AUC), Brier score, and their 95% confidence intervals (CIs). It assessed three performance dimensions: discrimination (AUC, sensitivity, specificity, positive predictive value, and negative predictive value), calibration (calibration curves, Hosmer–Lemeshow (H-L) test, and Brier score). Clinical utility was evaluated using decision curve analysis (DCA) to compare net benefits across risk thresholds. The optimal cutoff was determined using the Youden index (sensitivity + specificity –1), prioritizing high sensitivity. Sensitivity analysis Age was prioritized for sensitivity analysis because it is a clinically recognized independent risk factor for urolithiasis, yet no consensus exists regarding optimal stratification cutoffs. To assess age stability and appropriateness of stratification, restricted cubic spline (RCS) was used to characterize the association between age and urolithiasis risk. Threshold analysis identified data-driven age cutoffs, which were compared with predefined clinical cutoffs (30/60 years) to assess performance consistency. Using LASSO-selected predictors, 1,000 bootstrap resamples were performed to compare models with and without stratification at 30 and 60 years based on AUC, Brier score, net reclassification improvement (NRI) and integrated discrimination improvement (NDI) to confirm age retention. Subgroup analysis Subgroup analyses (sex, age groups, BMI, water intake, and family history) were conducted to explore predictor–risk association consistency. Interaction tests (p < 0.05) assessed differential associations across subgroups. Additionally, the temporal validation cohort and the external geographic validation cohort were used to assess generalizability. A naive benchmark model (primary care variables selected via the enter method) was constructed to evaluate incremental values. Predefined calibration strategy When predicted probabilities deviated from actual values, the temporal validation cohort was prioritized for calibration, as it better reflects temporal stability and aligns with the target Lao population. Results Baseline characteristics and predictor screening The study initially included 665 consecutive participants (encompassing dry and rainy seasons) as the training group. An additional 205 participants recruited from a later period formed the temporal validation cohort, and 300 participants constituted the external geographic validation cohort (Fig. S1). Urolithiasis prevalence was 23.0% (153/665) in the training set, 24.9% (51/205) in the temporal validation cohort, and 19.7% (59/300) in the external geographic validation cohort. Among the indicators across the three sets, only drinking history showed a significant difference (p = 0.009), whereas all other indicators (including urolithiasis prevalence) had no significant differences (all p > 0.05; Table S1). In the training set, subgroup comparisons between participants with and without urolithiasis showed significant differences in patient source, daily water intake, sodium consumption, and blood calcium (all p < 0.05; Table 1). 10.7717/peerj.21644/table-1Table 1Baseline demographic and clinical characteristics of participants.CategoriesStone (%)Non-stone (%)p value95% CITotal153 (23.0)512 (77.0)Age (y, IQRs)48 (21.0)45 (27.2)0.370HLΔMed = 1.00 (−1.00, 4.00)Age group (n, %)0.329–<3029 (19.0)121 (23.6)30–5990 (58.8)268 (52.3)≥ 6034 (22.2)123 (24.1)BMI (kg/m2, SD)23.5 ± 2.523.4 ± 2.80.908MD=0.03 (−0.47, 0.53)Sex (n, %)Male104 (68.0)304 (59.4)0.068OR = 1.45 (0.99, 2.13)Female49 (32.0)208 (40.6)Patient source (n, %)Rural108 (70.6)272 (53.1) <0.001 OR = 2.12 (1.43, 3.13)Urban45 (29.4)240 (46.9)Physical labor (n, %)Yes110 (71.9)347 (67.8)0.387OR = 1.22 (0.82, 1.81)Family history (n, %)Yes30 (19.6)81 (15.8)0.336OR = 1.30 (0.82, 2.07)Hypertension (n, %)Yes36 (23.5)119 (23.2)0.998OR = 1.02 (0.66, 1.56)Diabetes (n, %)Yes23 (15.0)68 (13.3)0.675OR = 1.16 (0.69, 1.93)Smoking history (n, %)Yes45 (29.4)140 (27.3)0.691OR = 1.11 (0.74, 1.65)Drinking history (n, %)Yes49 (32.0)161 (31.4)0.971OR = 1.03 (0.70, 1.51)Daily water intake (n, %) <0.001 –<500 ml67 (43.8)54 (10.5)500–1,500 ml71 (46.4)238 (46.5)>1,500 ml15 (9.2)220 (43.0)Sodium consumption (n, %) <0.001 –<5 g15 (9.8)165 (32.2)5–10 g92 (60.1)299 (58.4)>10 g46 (30.1)48 (9.4)Blood calcium (mg/dL, IQR)10.1 (1.2)9.5 (0.5) <0.001 HLΔMed = 0.60 (0.50, 0.70)Blood uric acid (mg/dL, IQR)5.9 (0.5)5.9 (0.6)0.803HLΔMed = 0.00 (−0.10, 0.10) Notes. Data are presented as mean ± standard deviation (SD), n(%), or median (interquartile ranges, IQR). Significant p values (<0.05) are shown in bold. CIconfidence intervals ORodds ratio for binary variables (with “No” category as reference). MDmean difference derived from independent-samples t-test for BMIHLΔMedHodges–Lehmann median difference. En dash (“–”) indicates no corresponding statistical value available. The 95% CI column presents Hodges–Lehmann median difference (HLΔMed) and its 95% CI for continuous variables; for detailed pairwise comparisons (percentage point differences with 95% CIs) of multi-level categorical variables, see Table S6. BMI was compared by independent t-test (MD with 95% CI). Non-normal continuous variables (age, blood calcium, blood uric acid) were analyzed by Mann–Whitney U test (HLΔMed with 95% CI). Binary variables were tested via χ2 test (OR with 95% CI); three-level categorical variables (age group, daily water intake, sodium consumption) were assessed via omnibus χ2 test. Ranges (min–max) for blood calcium and blood uric acid were 8.3–10.9 mg/dL and 4.2–7.3 mg/dL in the stone group, and 8.1–10.9 mg/dL and 3.9–7.5 mg/dL in the non-stone group, respectively. Baseline demographic and clinical characteristics of participants. Notes. Data are presented as mean ± standard deviation (SD), n(%), or median (interquartile ranges, IQR). Significant p values (<0.05) are shown in bold. CIconfidence intervals ORodds ratio for binary variables (with “No” category as reference). MDmean difference derived from independent-samples t-test for BMIHLΔMedHodges–Lehmann median difference. confidence intervals odds ratio for binary variables (with “No” category as reference). mean difference derived from independent-samples t-test for BMI Hodges–Lehmann median difference. En dash (“–”) indicates no corresponding statistical value available. The 95% CI column presents Hodges–Lehmann median difference (HLΔMed) and its 95% CI for continuous variables; for detailed pairwise comparisons (percentage point differences with 95% CIs) of multi-level categorical variables, see Table S6. BMI was compared by independent t-test (MD with 95% CI). Non-normal continuous variables (age, blood calcium, blood uric acid) were analyzed by Mann–Whitney U test (HLΔMed with 95% CI). Binary variables were tested via χ2 test (OR with 95% CI); three-level categorical variables (age group, daily water intake, sodium consumption) were assessed via omnibus χ2 test. Ranges (min–max) for blood calcium and blood uric acid were 8.3–10.9 mg/dL and 4.2–7.3 mg/dL in the stone group, and 8.1–10.9 mg/dL and 3.9–7.5 mg/dL in the non-stone group, respectively. For the training set, outliers were identified in three variables: 6 (0.90%) for BMI, 33 (4.96%) for blood calcium, and 29 (4.36%) for blood uric acid (Fig. S2). All these outliers fell within the normal physiological range or were only slightly abnormal; thus, no further handling was performed. No missing values were present in the final analytical dataset; therefore, no imputation procedures were performed. GVIF values of all final predictors were <1.5 (no multicollinearity; Fig. S3). LASSO regression with five-fold internal cross-validation (1,000 resamples) was used for variable selection in the training set. At the optimal λ value of 0.0074, the coefficients of BMI, physical labor, smoking history, and drinking history were reduced to zero, ultimately retaining 10 predictors for further analysis (Fig. S4). Identification of independent predictors by regression analysis Univariate analysis first identified patient source, daily water intake, sodium consumption, and blood calcium levels) as significant predictors of urolithiasis, which were then included in multivariate logistic regression alongside age (p = 0.401) and sex (p = 0.056)—two variables with potential clinical relevance to urolithiasis, despite not reaching significance in the univariate analysis (Table 2). Urban residence remained a protective factor (OR = 0.52, 95% CI [0.31–0.86], p = 0.011). Female sex was also protective (OR = 0.59, 95% CI [0.35–0.99], p = 0.045). Water intake demonstrated a dose–response effect, with intakes of 500–1,500 mL/day (OR = 0.16, 95% CI [0.09–0.30], p < 0.001) and >1,500 mL/day (OR = 0.02, 95% CI [0.01–0.05], p < 0.001) showing protective effects. Sodium intake also exhibited a dose–response relationship, with 5–10 g/day (OR = 4.58, 95% CI [2.21–9.50], p < 0.001) and >10 g/day (OR = 21.90, 95% CI [9.50–52.98], p < 0.001) associated with increased risk. Blood calcium level was an independent risk factor (OR = 9.12, 95% CI [5.59–4.88], p < 0.001). For age, stratified analysis yielded ORs of 1.54 (95% CI [0.83–2.87], p = 0.173) for 30–59 years and 1.13 (95% CI [0.54–2.35], p = 0.743) for ≥60 years; although neither subgroup demonstrated a statistically significant association, age was retained in the model given its well-documented clinical relevance (Table 3). Development of the predictive nomogram Based on multivariate regression results, six independent predictors—patient source, daily water intake, sodium consumption, sex, age, and blood calcium levels—were used to construct a clinical prediction model. The model was visualized as a nomogram, in which scores for each factor were summed to estimate an individual’s urolithiasis probability (Fig. 1). 10.7717/peerj.21644/table-2Table 2Univariate logistic regression analysis of the training set (n = 665).VariableOR (95% CI)p valueAge1.00 (0.99, 1.02)0.40130–59––≥ 60––Sex0.69 (0.47, 1.01)0.056Hypertension1.02 (0.66, 1.56)0.941Patient source0.47 (0.32, 0.70) <0.001 Daily water Intake0.24 (0.18, 0.32) <0.001 500–1,500––>1,500––Sodium consumption3.22 (2.33, 4.45) 0.038 5–10––>10––Blood calcium5.41 (3.45, 7.80) <0.001 Family history0.78 (0.46, 1.30)0.337Diabetes1.16 (0.69, 1.93)0.580Blood uric acid0.97 (0.70, 1.31)0.835 Notes. ORodds ratio CIconfidence interval For categorical variables with >2 levels, the lowest level was used as reference (ref. = 0); Significant p values (<0.05) are shown in bold. Univariate logistic regression analysis of the training set (n = 665). Notes. ORodds ratio CIconfidence interval odds ratio confidence interval For categorical variables with >2 levels, the lowest level was used as reference (ref. = 0); Significant p values (<0.05) are shown in bold. Validation and performance of the nomogram Sensitivity analyses were conducted to verify the stability of age as a predictor and the appropriateness of its categorization. RCS confirmed a nonlinear association for age (p = 0.025) and blood calcium (p = 0.185; Fig. S5). Threshold analysis identified optimal cutoffs at 44.8/65 years. Event rates with data-driven cutoffs (45/65 years) were 21.4%, 26.7%, and 15.9%, and conventional cutoffs (30/60 years) were 19.3%, 25.1%, and 21.6%, respectively. Since the two strategies yielded similar distributions of incidence and predictive performance (Table S2), 30/60 cutoffs were adopted for better clinical applicability. Internal validation using 1,000 bootstrap resamples compared a five-variable LASSO model (including significant factors) with a six-variable model that incorporated the 30/60 age stratification. The two models showed minimal differences, with AUCs of 0.897 vs. 0.898, identical Brier scores of 0.103 both, sensitivity of 0.564 vs. 0.562, identical specificity of 0.934, and overlapping 95% CIs (Table S3). NRI (0.002) and IDI (0.003) were close to zero (CIs crossing zero). These findings confirm that the inclusion of clinically meaningful age stratification (30/60 years) does not compromise model performance, supporting its retention in the final model. 10.7717/peerj.21644/table-3Table 3Multivariate logistic regression analysis of the training set (n = 665).VariableOR (95%CI)p valueCoefficientIntercept−22.38Age (y)30–591.54 (0.83, 2.87)0.1730.43≥ 601.13 (0.54, 2.35)0.7430.12Sex0.59 (0.35, 0.99) 0.045 −0.52Patient source0.52 (0.31, 0.86) 0.011 −0.66Daily water Intake (mL/d)500–1,5000.16 (0.09, 0.30) <0.001 −1.80>1,5000.02 (0.01, 0.05) <0.001 −3.75Sodium consumption (g/d)5–104.58 (2.21, 9.50) <0.001 1.52>1021.90 (9.50, 52.98) <0.001 3.09Blood calcium (mg/dL)9.12 (5.59, 14.88) <0.001 2.21 Notes. ORodds ratio CIconfidence interval For categorical variables with multiple subgroups, reference categories are defined as: age <30 years for age grouping, male for sex, rural residence for patient source, daily water intake <500 mL/d for water intake, sodium consumption <5 g/d for sodium intake. Blood calcium was included as a continuous variable with original unit (mg/dL). Sex was coded as 1 = male and 2 = female; patient source was coded as 0 = rural and 1 = urban. Daily water intake and sodium consumption were entered as categorical dummy variables in regression analysis. Age groups were incorporated directly as categorical strata without fixed numerical coding. Significant p values (<0.05) are shown in bold font. Multivariate logistic regression analysis of the training set (n = 665). Notes. ORodds ratio CIconfidence interval odds ratio confidence interval For categorical variables with multiple subgroups, reference categories are defined as: age <30 years for age grouping, male for sex, rural residence for patient source, daily water intake <500 mL/d for water intake, sodium consumption <5 g/d for sodium intake. Blood calcium was included as a continuous variable with original unit (mg/dL). Sex was coded as 1 = male and 2 = female; patient source was coded as 0 = rural and 1 = urban. Daily water intake and sodium consumption were entered as categorical dummy variables in regression analysis. Age groups were incorporated directly as categorical strata without fixed numerical coding. Significant p values (<0.05) are shown in bold font. Lao population urinary stone prediction model nomogram. Six variable axes are included (sex, age group, patient source, daily water intake, sodium consumption, and blood calcium). The regression equation is: Logit (P) = - 22.38 + 0.43 × (Age 30 –59) + 0.12 × (Age ≥ 60) 0.52 × (Sex: female = 1) –0.66 × (Patient Source: urban = 1) –1.80 × (Water 1) –3.75 × (Water 2) + 1.52 × (Sodium 1) + 3.09 × (Sodium 2) + 2.21 × (Blood Calcium). In this nomogram, daily water intake and sodium consumption are divided into three tiers: water intake (Water): 0 = <500 mL/d, 1 = 500–1,500 mL/d, 2 = >1,500 mL/d; sodium consumption (Sodium): 0 = <5 g/d, 1 = 5–10 g/d, 2 = >10 g/d. Each category on every axis is assigned a specific point value. Summing the points from all variables yields a total score, which corresponds to the predicted probability of urolithiasis. Detailed variable coding rules and reference categories are presented in the notes of Table 3. Subgroup analysis (Fig. S6) showed a significant sex interaction (p = 0.020; OR = 4.35 (male) vs. OR = 1.96 (female)). No significant interactions were found for age (OR: 3.12–4.31, p = 0.453), water intake (OR: 2.45–4.93, p = 0.087), BMI (p = 0.203), or family history (p = 0.970). Collectively, these analyses confirm that the nomogram’s key predictors are stable, with sex being the only factor showing differential risk effects. The nomogram showed satisfactory discriminative ability, with an AUC of 0.898 (95% CI [0.892–0.901]) in the training set. Its generalizability was further assessed in two independent validation datasets. As regards discriminative ability, the nomogram had an AUC of 0.882 (95% CI [0.829–0.936]) in the temporal validation cohort (Fig. 2A) and 0.907 (95% CI [0.866–0.948]) in the external geographic validation cohort (Fig. 2B). Notably, Platt calibration did not affect the AUC in either set (Table S4). Comparison of predictive performance between the final nomogram and a naive model. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values are shown for (A) the temporal validation cohort and (B) the external geographic validation cohort. Scarlet solid line: final nomogram (sex, age group, patient source, daily water intake, sodium consumption, and blood calcium). Deep blue solid line: naive model (age, sex, patient source, physical labor, daily water intake, and sodium consumption). The final model exhibited favorable discriminative and calibration abilities across both external validation cohorts (Fig. 3), supporting direct use of the original nomogram without additional calibration (Table S4). Its discriminative ability was demonstrated, indicated by AUC values exceeding 0.880 in both sets (the temporal validation cohort, 0.882; external geographic validation cohort, 0.907), exceeding the reference naive model (AUC: 0.799–0.817) and remaining stable with minimal improvement from additional calibration. Without extra calibration, it also met the “well-calibrated” criteria, including H-L p value > 0.05 (suggesting no significant deviation between the predicted and actual outcomes), calibration slopes close to 1, and low Brier scores (minimal prediction error), with additional calibration only reducing Brier scores by 0.001. In contrast, the reference naive model required calibration to fix a suboptimal initial fit (H-L p = 0.004 in the external geographic validation cohort), further indicating distinct differences in baseline performance between the two models. Comparison of calibration curves before and after calibration for both the final model and the naive model. Subgraph (A) corresponds to the temporal validation cohort, and subgraph (B) represents the external geographic validation cohort. The diagonal dashed line denotes ideal perfect calibration. Scarlet solid line: final nomogram (pre-recalibration); deep blue solid line: naive baseline model (pre-recalibration); pale red dashed line: final nomogram (post-recalibration); pale blue dashed line: naive baseline model (post-recalibration). Curves overlapping closely with the diagonal line reflect strong concordance between model-predicted risks and observed clinical outcomes. For clinical utility, DCA showed that the nomogram achieved positive net benefit within the threshold probability range of 10%–50% in the training, temporal validation, and external geographic validation cohorts. Its net benefit values were higher than those of the naive model, treat-all and treat-none strategies. (Fig. 4; detailed metrics in Table S5), with optimal thresholds of 0.208 (the training cohort), 0.204 (temporal validation cohort), and 0.198 (external geographic validation cohort). Within the clinically relevant 10%–30% threshold range, the nomogram maintained stable positive net benefit (with mild downward trends: 0.199 to 0.134 in the temporal validation cohort, 0.152 to 0.106 in the external geographic validation cohort). Even at a higher 40% threshold, net benefit remained positive (temporal validation cohort, 0.101; external center validation cohort, 0.093). These results demonstrate the stable clinical utility of the uncalibrated nomogram across training and validation sets, supporting its potential for practical urolithiasis risk assessment. Decision curve analysis (DCA) compared the uncalibrated/calibrated model. Net clinical benefit for urolithiasis risk prediction is plotted across threshold probabilities ranging from 0%–50%. Curve definitions: scarlet solid line = uncalibrated final nomogram; deep blue solid line = uncalibrated baseline naive model; pale red dashed line = recalibrated final nomogram; pale blue dashed line = recalibrated baseline naive model; black dashed line = treat-all strategy; gray dashed line = treat-none strategy. Discussion Urolithiasis arises from a complex interaction of genetic, environmental, dietary, lifestyle, metabolic, and disease-related factors (Akram et al., 2024; Harrell Jr, 2026). This study translates epidemiological evidence into a practical tool for low-resource settings. It supports health promotion and helps primary care teams prevent noncommunicable diseases (NCDs). This study reported a urolithiasis prevalence of 23.0% among Lao urological inpatients, consistent with Southeast Asian regional trends, providing the first contemporary epidemiological dataset on urinary stone disease in Laos. It addresses a key knowledge gap via systematic analysis of sociodemographic and laboratory predictors, offering insights for targeted prevention in this understudied regional population. The study’s major contribution lies in developing Laos’ first urolithiasis risk prediction nomogram using rigorous methods: multicollinearity assessment and LASSO regression filtered nonessential variables (ensuring a simple model and less bias), whereas internal resampling and external validation confirmed stable performance—critical for resource-limited settings. Additionally, although prespecified imputation strategies were established for potential missing data scenarios, all sporadic transcription gaps identified during database construction were fully recovered via manual cross-checking of original paper records. Consequently, no statistical imputation was required, ensuring the authenticity of the original clinical data for model development. Regarding variable importance (Fig. S7), the included factors balance modifiable risks (rural residence and water/sodium intake) and core demographic–metabolic factors (age, sex, and blood calcium). All variables are accessible in primary care, avoiding reliance on advanced diagnostics. These modifiable exposures illustrate the interplay between human beings and their physical (climate and water access), social (rural labor norms), and behavioral environments, underscoring the wider environmental determinants of urolithiasis emphasized in contemporary public health frameworks. Notably, the nomogram’s inherent favorable calibration supports direct clinical use, and age stratification by the familiar 30/60-year cutoff enhances usability for primary care providers (Zhu et al., 2024; Chinese Urological Association et al., 2024). This design also focuses on modifiable factors, helping the nomogram guide targeted interventions. It meets Laos’ specific need for practical, low-cost risk assessment, given that its NCD prevention strategies are still underdeveloped. Contrary to previous reports (Zhu et al., 2024; Okita et al., 2020), no significant associations were found between urolithiasis and age, BMI, or urological family history (all p > 0.05). Notably, although age was not significant in this study, it is a well-recognized clinical risk factor for urolithiasis and was therefore retained in the model (Okita et al., 2020). This divergence may reflect Laos’ concentrated age distribution—67.1% of participants aged 30–65 years, which is consistent with the preset age stratification (30/60 years) and covers the disease’s peak incidence—as well as context-specific factors, underscoring the need for localized research. This aligns with the results of studies from China (Aizezi et al., 2022), USA (Katz et al., 2021), and Germany (Siener et al., 2022), highlighting a high burden on working-age adults (likely due to strenuous labor and inadequate water intake) (Yao et al., 2025). Although historical data show the association of male predominance with physiology (Akram et al., 2024), our sex distribution (61.3% (male) vs. 38.7% (female)) aligns with this trend—though the narrowing sex gap reflects emerging epidemiological shifts (GBD 2021 Urolithiasis Collaborators, 2024), and multivariate analysis further confirmed that women had lower urolithiasis risk than men (OR = 0.59), highlighting the need for sex-sensitive prevention tailored to this evolving pattern. Our subgroup analysis further identified a significant sex interaction in the model’s predictive effect (p = 0.020), with men exhibiting a much higher urolithiasis risk (OR = 4.35) than women (OR = 1.96). This finding aligns with the results of large Chinese study by Zhang et al. (2021), which reported that men accounted for 68.2% of patients with urinary stone—reflecting a consistent sex disparity across regional populations, including Laos (Sayasone et al., 2004). Physiologically, this difference may stem from two key, mutually reinforcing factors: first, men typically have higher urinary calcium and oxalate excretion than women (Akram et al., 2024), which raises urine solute concentration (a key trigger for stone formation). Second, this physiological trait is amplified by higher rates of strenuous labor (68.7% of rural participants) and lower water intake (18.1% < 500 mL/day) among Lao men, further increasing urinary solute concentration. Notably, stone composition analysis was not performed to confirm the dominant stone type, which should be addressed in future research to refine this mechanistic understanding. Clinically, this interaction underscores our model’s value for prioritizing male risk assessment in resource-limited Lao inpatient settings. Additionally, urban residents had a 48% lower urolithiasis risk than rural residents (OR = 0.52)—likely due to rural Laos’ inadequate medical access, harsh climate, high strenuous labor (68.7%), and low water intake (18.1%). Water intake showed a clear protective dose effect: those consuming >1,500 mL/day had drastically reduced risk (OR = 0.02), aligning with evidence that intake of more fluid lowers stone risk by diluting urine (Wang, Zhang & Wei, 2021). Although guidelines recommend >3 L/day, targeted promotion for rural laborers is challenging (Geraghty et al., 2023). Multivariate results identified high sodium intake and elevated serum calcium as significant risk factors. However, these markedly elevated odds ratios should be interpreted with considerable caution, as the inpatient-only enrollment likely introduces spectrum bias and hospital-based selection bias that may inflate observed effect sizes. Such exaggeration may overestimate true associations in community populations, warranting verification in population-based cohorts. Nevertheless, these findings are biologically plausible: excess sodium increases urinary calcium excretion and facilitates stone formation (Ferraro et al., 2020), while even modest blood calcium elevations may indicate altered homeostasis. Whether these associations extend to primary care settings remains to be prospectively evaluated. Sodium reduction and calcium monitoring may merit consideration, but their population-level impact requires further investigation. Unlike previous studies (Skolarikos et al., 2024; Yang et al., 2023), no significant associations were found between diabetes, serum uric acid, or alcohol consumption and urinary stone incidence (all p > 0.05), possibly due to low local uric acid stone prevalence. Stone composition analysis was not performed because of technical constraints. EAU and AUA guidelines recommend future studies incorporate infrared spectroscopy to refine prevention strategies (Geraghty et al., 2023; Skolarikos et al., 2024). Notably, while the final model maintained consistent performance in the external geographic validation cohort, cross-regional generalization challenges were observed when referencing data from northwest China (north of the Yellow River), where urolithiasis incidence is generally <14% of urological inpatients in the same period (Wang et al., 2022)—linked to local cold climate and sufficient water intake. This contrasts sharply with Laos (Southeast Asia, hot/humid, widespread insufficient water intake). Beyond these environmental differences, the Lao training/temporal validation cohorts and the Chinese external geographic validation cohort differ substantially in healthcare infrastructure (centralized tertiary system with standardized referral vs. resource-constrained fragmented care) (Wang et al., 2022; Ministry of Lao Health PDR, 2017), demographics (predominantly Han ethnicity with distinct dietary patterns vs. multi-ethnic younger population with agrarian diets) (Liu et al., 2018), and environmental exposures (water source and quality, dietary mineral intake, and heat exposure) (Liu et al., 2018; Wang et al., 2022). These multidimensional differences suggest that caution is warranted when generalizing the nomogram to populations with markedly different epidemiological backgrounds, and that context-specific validation should precede broader implementation. Notably, our model exhibited steady inherent calibration without additional calibration (Table S4), simplifying its practical application in Lao inpatient settings where complex statistical adjustment is often unfeasible. The model’s stable discriminative ability (consistent AUC across training and validation sets) paired with its inherent calibration ensures that risk estimates are accurate for clinical use—critical for guiding screening decisions in resource-limited primary care. By integrating urology, nutrition science, and rural public health, this study exemplifies the transdisciplinary collaboration required to reduce NCD burden in low-resource settings. Based on model performance and risk factors, the nomogram’s clinical use depends on two key links: threshold determination and on-site implementation. DCA results (Table S5) guide threshold setting, and the nomogram delivered positive net benefit across 10%–50% thresholds (compared with the naive model and “treat-all”/”treat-none” strategies) and remained stable in the 10%–30% clinical core range. Although Youden index cutoffs varied slightly (0.198–0.208), we prioritized opting for a 0.20 general threshold. This balances detection and reduces missed cases, avoiding set-specific inconsistencies (Youden optimizes single-set balance; DCA reflects real decision value). Collectively, this outlines a practical path for Laos: with urologist support, local teams use the nomogram for rural screening (simple inquiries such as residence, diet, and basic blood tests) to identify high-risk patients. These patients can then be referred for further management. It meets low-resource needs, aligning with the Ministry of Lao Health PDR (2017) and WHO-recommended clinical tools for low-resource settings (World Health Organization, 2021). Urologists can use this tool for district-level screenings by training health workers to use portable calcium meters and conduct dietary surveys—predicted risks are calculated via the validated calibration formula to ensure accuracy. High-risk cases can then be referred for ultrasound confirmation, which streamlines triage and makes better use of specialists. This workflow develops a low-cost primary healthcare-based screening model that enhances urological service capacity and supports the WHO’s universal health coverage objectives in resource-limited settings (World Health Organization, 2021). Our nomogram directly contributes to national and global health goals and supports Sustainable Development Goal 3.4 (UN General Assembly, 2015). In summary, this integrated approach enables early urological detection in remote communities, strengthening preventive capacity, and promoting equity in stone disease management. Furthermore, this tool has potential reference value for other Southeast Asian regions with similar geography, culture, and population characteristics, yet cross-regional adaptation and formal verification are still required before widespread implementation. However, this study has several limitations. First, the use of hospital inpatient samples may introduce selection and spectrum biases, limiting generalizability to the broader Lao community. Second, the retrospective secondary analysis design and the predefined ≥80% data completeness threshold theoretically carry a minor risk of selection bias, although no cases required exclusion because of missing core variables after exhaustive verification of original records. Third, important data were missing: stone composition (critical for tailored prevention), genetic and metabolic markers, and objective diet records. Fourth, the water and sodium intake were assessed using a 24-h dietary recall. While this method is widely used in epidemiological studies (DeKeyzer et al., 2015), formal reliability and validity verification targeting the Lao local population has not been performed. Compared with urinary sodium measurement, self-reported sodium and fluid intake carries inherent recall bias and commonly underestimates true intake (McLean et al., 2019; Cogswell et al., 2018; Mercado et al., 2015; Durán-Cabral et al., 2025). Retrospective spot urine testing was not feasible because of the unavailability of stored specimens. The broad range of sodium intake and high water intake while reflecting local dietary heterogeneity may introduce measurement variability at category boundaries. Fifth, the nomogram was developed for Lao populations and may need recalibrations in other regions. Sixth, because all validations were restricted to hospitalized cohorts, prospective pilot implementation in community-based primary care settings has not been conducted. The present nomogram was derived and validated solely among inpatients as the first stage of a staged research roadmap, so it is currently suitable mainly for inpatient risk stratification in local clinical practice. Future prospective studies should incorporate community-based populations and multicenter designs, alongside stone composition analysis, to further validate its utility and assess its generalizability. Conclusions This study presents the first epidemiological data and validated nomogram for urolithiasis risk screening in Laos. Key predictors included age, sex, rural residence, low water intake, high sodium consumption, and elevated blood calcium. Currently, the nomogram is suitable mainly for inpatient decision-making in Laos, where a 0.20 clinical threshold—supported by decision curve analysis—may offer practical guidance for clinicians. However, its application in community-based primary care remains contingent upon prospective verification. Future studies should prioritize cross-regional validation in other Southeast Asian populations, explore integration with electronic medical records, and incorporate stone composition and environmental risk factor analyses to inform targeted prevention strategies. Supplemental Information Questionnaire Participant flowchart for model training and validation datasets Flow diagram of study population selection based on the TRIPOD guidelines. The same inclusion and exclusion criteria were applied across all datasets, including the external validation cohorts. (A) Temporally independent validation set (Vientiane Military Hospital, November 2024–February 2025): initial n = 183, 23 cases excluded per the training set criteria, final n = 160. (B) External-center validation set (Xijing Hospital, Xi’an, April–June 2025): initial n = 260, 36 cases excluded per the training set criteria, final n = 224. Baseline characteristics Outlier distribution of BMI, blood calcium, and blood uric acid in the training set Box plots showing the distribution of three variables (BMI, blood calcium, blood uric acid) in the training set, with outliers identified via the interquartile range (IQR) method marked in red dots. The number and proportion of outliers for each variable are as follows: BMI (6 cases, 0.90%), blood calcium (33 cases, 4.96%), and blood uric acid (29 cases, 4.36%). Generalized variance inflation factor (GVIF) analysis for multicollinearity assessment A bar chart visualizing the GVIF values of all predictor variables in the multivariate regression model, with the red dashed line indicating GVIF = 1. The GVIF values of all the variables in the chart were less than 1.5, indicating no significant multicollinearity. Generalized variance inflation factor (GVIF) analysis for multicollinearity assessment. Variable selection using LASSO regression (A) Coefficient paths of each variable. The red and black dashed vertical lines correspond to the optimal λ values (λ .min and λ .1se, respectively). Numbers above indicate the number of non - zero coefficients. (B) Binomial deviance versus ( λ). Nonlinear associations between age, blood calcium levels, and urolithiasis Scatter points: Individual participant data (gray/black dots represent observed cases of urolithiasis presence/absence). Red curve: Restricted Cubic Spline (RCS)—derived predicted probability of urolithiasis presence (significant nonlinear trend for age, p = 0.025; non - significant trend for blood calcium, p = 0.185). Blue dashed line: Reference baseline for probability comparison. Age-grouping strategies versus continuous-age RCS reference Performance comparison (Bootstrap, 1000 replications) between 5-variable LASSO and 6-variable age-stratified LASSO models Subgroup analysis of the association between exposure factor and urolithiasis Horizontal blue line: 95% Confidence Interval (95% CI) for the Odds Ratio (OR). Blue square: Point estimate of the OR (representing the strength of association in each subgroup). Vertical red line: Reference line (OR = 1, indicating no association). P - interaction: p value for the interaction test (evaluating whether the association differs across subgroups). Performance metrics Model performance and net benefit Odds ratios of risk factors for urolithiasis susceptibility Blue dot: Point estimate of the Odds Ratio (OR) for each variable (representing the magnitude of association with urolithiasis risk). Horizontal blue line: 95% Confidence Interval (95% CI) of the OR (reflecting the uncertainty of the estimate). Vertical red dashed line: Reference line (OR = 1, indicating no association between the variable and urolithiasis risk). Percentage point differences (stone group % minus non-stone group %) with Wilson score 95% CIs Percentage point differences between the urolithiasis group and non-stone control group for all categorical predictors, with corresponding 95% confidence intervals calculated via the Wilson score method for each variable. TRIPOD checklist: prediction model development and validation STROBE checklist De-identified raw clinical datasets of study participants De-identified raw clinical dataset of enrolled participants for the urolithiasis prediction model, covering baseline demographics, laboratory metabolic parameters and stone diagnostic information; all sensitive personal identifiers have been removed.