Development and validation of a predictive model for hypothalamic obesity following craniopharyngioma resection Authors: Qianshuai Li, Linlin Zhao, Fengjuan Huang, Xuyang Gong, Mengxing Pan, Xinjing Liu, Tianfang Wang, Xulong Li, Yijing Xie, Ankang Gao, Yong Zhang, Yanyan Zhao Published: 2026 ABSTRACT ObjectiveTo analyze the risk factors for hypothalamic obesity (HO) after craniopharyngioma surgery, construct a predictive model and verify its efficacy.MethodsA retrospective analysis was conducted on 136 patients diagnosed with craniopharyngioma at The First Affiliated Hospital of Zhengzhou University between September 2018 and February 2024. Clinical and follow-up data were collected and split into training and validation cohorts (7: 3) randomly and divided into HO and non-HO groups post-surgery. Logistic regression was used to analyse risk factors for postoperative HO, and a prediction model was constructed and its efficacy was verified.ResultsIn the training cohort, the preoperative body mass index (BMI), preoperative serum sodium, preoperative uric acid levels, postoperative calcium levels, the probability of intraoperative blood loss ≥ 100 ml and preoperative hypothalamic injury and T-type tumors were higher in HO group. Small tumors (< 5 cm³) were less frequent in HO group (all P < 0.05). Logistic regression analysis showed that preoperative high BMI and intraoperative blood loss ≥ 100 ml were independent risk factors for postoperative HO, while older age at onset and smaller tumor size were protective factors (all P < 0.05). The model developed from the training cohort showed that the area under the receiver operating characteristic curve (AUC) was 0.865(95%CI of 0.793 ~ 0.937), with a sensitivity of 79.5% and a specificity of 80.8%, the validation cohort had an AUC of 0.864(95%CI:0.744–0.983), with a sensitivity of 90.9% and a specificity of 72.2%. The calibration curve and Hosmer-Lemeshow test demonstrated that the predicted value of the model was in agreement with the observed values in both the training and verification cohorts (P > 0.05).ConclusionPreoperative high BMI and intraoperative blood loss ≥ 100 ml are independent risk factors for HO after craniopharyngioma surgery. A predictive model based on preoperative BMI, age at onset, intraoperative blood loss, and tumor size shows potential for predicting the risk of postoperative HO, aiding in early identification for high-risk patients.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12902-026-02322-x. Introduction Craniopharyngioma (CP) is a rare embryonal tumor occurring in the sellar region or near the sella turcica. Although it is benign in tissue, it is located deeply and adjacent to important structures such as the hypothalamus, pituitary and optic nerve / optic chiasm. It often leads to the decline of hypothalamic-pituitary function after surgery, and seriously damages the quality of life of affected patients [1]. Hypothalamic obesity (HO) refers to the syndrome of hyperphagia and significant weight gain caused by structural or functional impairment of the hypothalamic energy homeostasis regulation system. Previous studies have shown that the incidence of secondary HO after craniopharyngioma was from 51% to 78% and 27% to 55.6% in children and adults respectively [2–4]. High obesity rate can cause the risk of metabolic syndrome and cardiovascular disease, which will seriously affect the quality of life of patients [5]. At present, the association of hypothalamic involvement and HO has been confirmed, but the effects of other factors such as high preoperative body mass index (BMI), surgical mode and postoperative radiotherapy on postoperative hypothalamic obesity after craniopharyngioma are still controversial [6–7]. This indicates that a single risk factor is insufficient for clinical assessment. Therefore, this study aimed to analyze risk factors for HO after craniopharyngioma surgery, construct a multivariable predictive model, and perform internal validation as a preliminary step toward early identification of high-risk patients. Methods Research design and participants A flowchart of the study was presented in Fig. 1. In this study, 136 patients diagnosed with craniopharyngioma at the First Affiliated Hospital of Zhengzhou University between September 2018 and February 2024 were retrospectively analyzed. The inclusion criteria were as follows: (1) preoperative imaging revealed a lesion in the sellar region, which was confirmed as craniopharyngioma (CP) on postoperative pathological examination, encompassing both histological subtypes: adamantinomatous craniopharyngioma (ACP) and papillary craniopharyngioma (PCP); The exclusion criteria included: (1) the clinical data were incomplete and the follow-up data were missing; (2) recurrent tumors; (3) patients with other endocrine disorders known to cause secondary obesity (e.g., hypercortisolism) or with other obesity-associated conditions (e.g., polycystic ovary syndrome). Fig. 1Flowchart of study participant enrollment Flowchart of study participant enrollment This study adheres to the TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) guidelines for reporting prediction model studies. Data completeness criterion For the purpose of this study, patient data were considered “complete” only if they contained the minimum information set necessary to: (1) Preoperative BMI and sufficient postoperative data (a measured BMI or definitive clinical documentation) to determine weight change within the first year. (2) Basic clinical descriptors: age at surgery, key tumor features from MRI, and intraoperative blood loss. The presence of all these elements was mandatory for inclusion in the analytical cohort, as their absence would preclude the reliable assessment of the outcome or the patient’s baseline clinical profile. Patients who did not meet the above completeness criterion were excluded from the analysis, corresponding to a complete-case analysis approach. Among the 284 patients initially screened, 148 were excluded due to incomplete data, yielding a final analytical cohort of 136 patients. Variable selection Predictors for univariate analysis were selected a priori based on clinical rationale and previously reported associations, including age, BMI, hypothalamic involvement, tumor volume, and extent of resection. Variables with biological relevance to hypothalamic obesity were prioritized, and those with limited prior evidence were also considered for exploratory analysis. Handling of continuous variables Age and preoperative BMI were analyzed as continuous variables. Tumor volume was analyzed as categorical (small vs. non‑small, using a cutoff of 5 cm³). Intraoperative blood loss was analyzed as categorical (< 100 mL vs. ≥100 mL). All other continuous variables (e.g., serum sodium, uric acid, calcium) were examined in their original continuous form in univariate analyses without transformation. Clinical data were retrieved and collected through the Hospital Information System (HIS), including the following: (1) patient characteristics: age, sex, height, weight, and BMI. (2) perioperative clinical and laboratory data: intraoperative blood loss, preoperative and postoperative (day 1) measurements, including biochemical profiles and pituitary hormone panels. (3) imaging data: the characteristics of pituitary MRI plain scan and enhanced scan were assessed by two radiologists with intermediate or higher professional titles, including tumor location classification (QST classification, Q-type: infradiaphragmatic; S-type: suprasellar-subarachnoid; T-type: tuberoinfundibular) [8], tumor volume (①small: tumor volume<5 cm³②non-small: tumor volume≥5 cm³), tumor volume was determined by the maximal diameters in 3 dimensions using the following formula: volume = anteroposterior × craniocaudal × transverse/2 [9], grade of hypothalamic involvement (according to Puget classification, Grade 0, no hypothalamic involvement; Grade 1, tumor abuts or compresses the hypothalamus; Grade 2, significant hypothalamic damage or unrecognizable floor of the third ventricle) [10], the degree of hydrocephalus (assessed by the ratio of ventricular diameter to biparietal diameter: the normal value is less than 25%; mild hydrocephalus is defined as 26%−40; moderate hydrocephalus as 41%−60%; and severe hydrocephalus as greater than 60%) [11], presence or absence of tumor calcification, and whether the pituitary gland and pituitary stalk are visible. (4) surgery-related data: degree of tumor resection (gross total resection: no residual tumor on postoperative MRI review; subtotal resection: tumor resection ≥ 80%; partial resection: tumor resection < 80%) [12]. (5) follow-up data: the height and weight of patients were followed up through outpatient or telephone. (6)Serum calcium levels were adjusted for albumin concentration using the standard formula: Adjusted calcium (mmol/L) = Total serum calcium (mmol/L) + 0.02 × [40-Serum albumin (g/L)]; Based on the diagnostic criteria for HO, which integrates domestic and international research, patients were defined as meeting the following: significant weight gain within one year after tumor resection was characterized by a clinical increase in BMI (≥ 5% for adults; ≥1.0 SDS for pediatric/adolescent patients), with a resultant BMI ≥ 28 kg/m² in adults or ≥ 95th percentile (P95) in age- and sex-matched pediatric/adolescent patients, along with at least one clinical manifestation of non-endocrine hypothalamic dysfunction [13–15]. Given the diagnostic criteria for HO specifies significant weight gain within the first postoperative year, the primary analysis focused on weight change observed during this period. For patients whose first postoperative BMI measurement occurred after 12 months, the occurrence of HO within the first year was adjudicated through a structured review of all available clinical records from the interim period (e.g., clinic notes, nursing records, discharge summaries). Patients were classified as HO if these records contained positive documentation (e.g., explicit notation of significant weight gain, new-onset hyperphagia, or a clinical diagnosis of HO) within the first 12 months. Conversely, if no such documentation was found upon comprehensive review, they were classified as not having developed HO within the first year for the purpose of this study. Patients with confirmed weight gain after 12 months were not classified as HO in this analysis. Postoperative BMI was calculated using the body weight and height measured at the first available follow-up visit after surgery. The timing of this initial assessment varied among patients (range: 2–67 months) due to individual clinical circumstances. For the diagnosis of HO, this postoperative BMI value served as the key comparative endpoint against the preoperative BMI. Non-endocrine hypothalamic dysfunction encompasses appetite/feeding abnormalities, sleep-wake cycle disturbances, affective disorders, cognitive-behavioral abnormalities. To objectively quantify the key components of non-endocrine hypothalamic dysfunction, a standardized assessment battery was employed: appetite change was assessed using a 100-mm Visual Analogue Scale, with a sustained increase of ≥ 20 mm from baseline defining hyperphagia; daytime sleepiness was screened with the Epworth Sleepiness Scale, where a total score ≥ 10 indicated excessive somnolence; global cognitive function was evaluated using the Mini-Mental State Examination (MMSE), with a score of < 27 defining cognitive impairment; affective disorders were measured using the Patient Health Questionnaire−9, with a score ≥ 10 suggesting clinically significant depressive symptomatology. The presence of clinically significant abnormalities in one or more of these domains supported the diagnosis [16–19]. To ensure a comprehensive and age-appropriate assessment, the evaluation of non-endocrine dysfunction integrated information from multiple sources, including standardized questionnaires (when applicable), proxy parental reports, and systematic review of clinical documentation. Subjects were classified into HO and non-HO groups. All data were randomly divided into training and validation cohorts (7: 3), with model development conducted on the training cohort. Assessment of pituitary function: ①Insufficient growth hormone (GH) secretion: In this study, without performing GH stimulation tests, we defined insufficient GH secretion as serum GH levels < 0.06 ng/mL(reference range: 0.06−5.0ng/mL via chemiluminescence assay). ②Central hypothyroidism was diagnosed based on free T4 level below the reference range combined with a low or normal TSH. ③Central Hypogonadism: evaluation of the hypothalamic-pituitary-gonadal (HPG) axis was restricted to patients who had chronologically surpassed the theoretical pubertal onset age (> 13 years for males, > 11 years for females): In premenopausal women, central hypogonadism was defined by oligomenorrhea or amenorrhea combined with low serum estradiol and inappropriately low or normal FSH and LH levels. Gonadotrophin deficiency in postmenopausal women was diagnosed by serum FSH and LH within premenopausal range. In men, central hypogonadism was defined as low serum testosterone in conjunction with low gonadotropins [20]. ④Central adrenocortical hypofunction: plasma cortisol (COR) < 5.0 µg/dL at 8:00 a.m. with normal or decreased adrenocorticotropic hormone (ACTH) levels [21]. Hyperprolactinemia assessment: defined as serum prolactin (PRL) levels exceeding the upper limit of normal reference range (> 17.7 µg/L in males; >29.2 µg/L in females). This was measured as a descriptive biochemical variable. In this study, preoperative prophylactic glucocorticoids were not utilized. Hormone levels measured on postoperative day 1 are presented descriptively, they were not used in isolation to establish long-term endocrine diagnoses. Statistical analysis Normally distributed continuous data were expressed as mean ± standard deviation (x̄±s) and compared between groups using independent samples t-test. Non-normally distributed continuous data were reported as median (Q1, Q3) and analyzed with Mann-Whitney U test. Categorical variables were presented as frequencies (percentages) [n (%)], with intergroup comparisons performed by χ² test or Fisher’s exact probability test. Independent risk factors for postoperative HO were identified through logistic regression analysis. The prediction model was developed and its performance validated using R-4.3.2 software, the points assigned to each predictor in the prediction model are automatically calculated by the R package based on the regression coefficients (β values), which inherently reflect the relative weight or contribution of each variable to the outcome risk in our dataset. Two-sided p < 0.05 was considered statistically significant. Results In the entire cohort of 136 patients, 66 were diagnosed with postoperative HO. Of these, 49 were diagnosed based on BMI measurements obtained within the first postoperative year, whereas the remaining 17 were diagnosed through structured review of clinical records (due to the first postoperative BMI measurement occurring after 12 months). As shown Tables 1 and 2 in the appendix, the training and validation cohorts were generally well balanced across most clinical and tumor characteristics, with statistically significant differences observed only in a limited number of baseline variables. Table 1Comparison of clinical characteristics of HO patients and non-HO patients in the training cohortVariablesPreoperative variablesPostoperative variablesHO group(n = 44)non-HO group(n = 52)P valueHO group(n = 44)non-HO group(n = 52)P valueMale (n, %)24(54.5%)32(61.5%)0.48924(54.5%)32(61.5%)0.489Age of onset (years)30(12.25, 51.50)40.50(20.25, 55.25)0.14530(12.25, 51.50)40.50(20.25, 55.25)0.145BMI (kg/m2)25.55(22.93, 28.63)23.20(18.98, 24.54)< 0.001*29.72 ± 4.34 23.50 ± 3.37< 0.001*K+ (mmol/L)4.22 ± 0.424.27 ± 0.340.4894.02 ± 0.513.87 ± 0.490.142Na+ (mmol/L)142.00(141.00, 143.83)141.00(139.00, 142.00)0.032*147.08 ± 11.34147.05 ± 8.220.989Cl− (mmol/L)104.60(102.00, 105.88)103.20(101.93, 104.00)0.068110.41 ± 10.95109.40 ± 7.920.601Ca2+ (mmol/L)2.35 ± 0.102.32 ± 0.100.0732.25 ± 0.142.16 ± 0.120.001*Corrected Ca2+ (mmol/L)2.26 ± 0.092.24 ± 0.080.1912.27 ± 0.192.18 ± 0.130.014*P3+ (mmol/L)1.39 ± 0.191.31 ± 0.220.0511.18 ± 0.331.17 ± 0.430.899Mg2+ (mmol/L)0.96(0.90, 1.04)0.97(0.92, 1.05)0.6510.93 ± 0.160.94 ± 0.150.752FBG (mmol/L)4.73(4.22, 5.20)4.57(4.03, 5.38)0.4606.17(4.70, 10.20)5.24(4.34, 9.35)0.137TG (mmol/L)1.71(0.86, 2.98)1.37(0.73, 2.74)0.513///TC (mmol/L)4.98 ± 1.274.64 ± 1.080.157///HDL-c (mmol/L)1.12(0.85, 1.43)1.21(0.98, 1.43)0.670///LDL-c (mmol/L)3.11 ± 1.112.73 ± 1.000.083///ALT (U/L)16(12.00, 30.00)19(13, 37.50)0.51523(13.00, 50.13)26.45(14.00, 45.58)0.991AST (U/L)19(17.00, 25.75)22.50(18.00, 28.75)0.30422(17.00, 53.25)23(16.00, 37.55)0.741GGT (U/L)19(13.00, 29.25)16.50(11.00, 29.75)0.42726(14.25, 54.00)21(11.25, 35.00)0.251Albumin (g/L)44.56 ± 2.7443.77 ± 3.500.22838.97 ± 6.2838.64 ± 4.340.764UA (µmol/L)319.10 ± 72.33278.73 ± 86.020.016*311.25 ± 107.32274.79 ± 115.480.1158:00 AM ACTH (pg/mL)28.88(13.45, 45.55)27.30(12.03, 47.52)0.5917.28(5.00, 11.38)7.59(4.09, 15.23)0.8778:00 AM COR (µg/dL)13.10(3.71, 17.15)11.44(4.41, 15.71)0.7054.00(1.87, 8.82)4.30(1.85, 9.05)0.971FT3 (pmol/L)4.39 ± 0.704.48 ± 0.890.6123.31(2.99, 3.96)3.43(2.97, 4.03)0.691FT4 (pmol/L)9.67 ± 2.998.89 ± 3.050.2118.75(5.76, 10.82)7.50(6.37, 10.74)0.607TSH (µIU/mL)1.77(1.23, 2.71)1.68(1.12, 3.03)0.9880.32(0.09, 1.60)0.32(0.11, 0.81)0.803GH (ng/mL)0.23(0.09, 0.53)0.23(0.10, 0.53)0.9060.27(0.05, 0.78)0.40(0.14, 1.08)0.098Prolactin (µg/L)32.95 (16.16,47.20)21.23 (11.04,44.62)0.13719.32(8.18,35.85)15.15 (6.03,38.59)0.633Hyperprolactinemia (n, %)28(63.6%)27(51.9%)0.24823(52.2%)19(36.5%)0.122Insufficient GH secretion (n, %)4(9.0%)6(11.5%)0.95513(29.5%)7(13.4%)0.053Hypogonadism (n, %)24(70.6%)[n = 34]23(51.1%)[n = 45]0.08133(97.1%)[n = 34]43(95.6%)[n = 45]1.000Hypothyroidism (n, %)12(27.2%)17(32.6%)0.56420(45.4%)31(59.6%)0.166Hypoadrenalism (n, %)13(29.5%)15(28.8%)0.94026(59.0%)30(57.6%)0.890Continuous variables with normal distribution were expressed as mean ± SD, non-normal data were expressed as median (interquartile range, IQR)BMI body mass index, FPG fasting blood glucose, HDL-c high-density lipoprotein cholesterol,LDL-c low-density lipoprotein cholesterol, TG triglycerides, TC total cholesterol, ALT alanine aminotransferase, AST aspartate aminotransferase, GGT gamma-glutamyl transferase, UA uric acid, ACTH adrenocorticotropic hormone, COR cortisol, FT3 free triiodothyronine, FT4 free thyroxine, TSH thyroid-stimulating hormone, GH growth hormone. Corrected Ca2+ = Total serum calcium + 0.02 × (40 − Serum albumin) Table 1Comparison of clinical characteristics of HO patients and non-HO patients in the training cohortVariablesPreoperative variablesPostoperative variablesHO group(n = 44)non-HO group(n = 52)P valueHO group(n = 44)non-HO group(n = 52)P valueMale (n, %)24(54.5%)32(61.5%)0.48924(54.5%)32(61.5%)0.489Age of onset (years)30(12.25, 51.50)40.50(20.25, 55.25)0.14530(12.25, 51.50)40.50(20.25, 55.25)0.145BMI (kg/m2)25.55(22.93, 28.63)23.20(18.98, 24.54)< 0.001*29.72 ± 4.34 23.50 ± 3.37< 0.001*K+ (mmol/L)4.22 ± 0.424.27 ± 0.340.4894.02 ± 0.513.87 ± 0.490.142Na+ (mmol/L)142.00(141.00, 143.83)141.00(139.00, 142.00)0.032*147.08 ± 11.34147.05 ± 8.220.989Cl− (mmol/L)104.60(102.00, 105.88)103.20(101.93, 104.00)0.068110.41 ± 10.95109.40 ± 7.920.601Ca2+ (mmol/L)2.35 ± 0.102.32 ± 0.100.0732.25 ± 0.142.16 ± 0.120.001*Corrected Ca2+ (mmol/L)2.26 ± 0.092.24 ± 0.080.1912.27 ± 0.192.18 ± 0.130.014*P3+ (mmol/L)1.39 ± 0.191.31 ± 0.220.0511.18 ± 0.331.17 ± 0.430.899Mg2+ (mmol/L)0.96(0.90, 1.04)0.97(0.92, 1.05)0.6510.93 ± 0.160.94 ± 0.150.752FBG (mmol/L)4.73(4.22, 5.20)4.57(4.03, 5.38)0.4606.17(4.70, 10.20)5.24(4.34, 9.35)0.137TG (mmol/L)1.71(0.86, 2.98)1.37(0.73, 2.74)0.513///TC (mmol/L)4.98 ± 1.274.64 ± 1.080.157///HDL-c (mmol/L)1.12(0.85, 1.43)1.21(0.98, 1.43)0.670///LDL-c (mmol/L)3.11 ± 1.112.73 ± 1.000.083///ALT (U/L)16(12.00, 30.00)19(13, 37.50)0.51523(13.00, 50.13)26.45(14.00, 45.58)0.991AST (U/L)19(17.00, 25.75)22.50(18.00, 28.75)0.30422(17.00, 53.25)23(16.00, 37.55)0.741GGT (U/L)19(13.00, 29.25)16.50(11.00, 29.75)0.42726(14.25, 54.00)21(11.25, 35.00)0.251Albumin (g/L)44.56 ± 2.7443.77 ± 3.500.22838.97 ± 6.2838.64 ± 4.340.764UA (µmol/L)319.10 ± 72.33278.73 ± 86.020.016*311.25 ± 107.32274.79 ± 115.480.1158:00 AM ACTH (pg/mL)28.88(13.45, 45.55)27.30(12.03, 47.52)0.5917.28(5.00, 11.38)7.59(4.09, 15.23)0.8778:00 AM COR (µg/dL)13.10(3.71, 17.15)11.44(4.41, 15.71)0.7054.00(1.87, 8.82)4.30(1.85, 9.05)0.971FT3 (pmol/L)4.39 ± 0.704.48 ± 0.890.6123.31(2.99, 3.96)3.43(2.97, 4.03)0.691FT4 (pmol/L)9.67 ± 2.998.89 ± 3.050.2118.75(5.76, 10.82)7.50(6.37, 10.74)0.607TSH (µIU/mL)1.77(1.23, 2.71)1.68(1.12, 3.03)0.9880.32(0.09, 1.60)0.32(0.11, 0.81)0.803GH (ng/mL)0.23(0.09, 0.53)0.23(0.10, 0.53)0.9060.27(0.05, 0.78)0.40(0.14, 1.08)0.098Prolactin (µg/L)32.95 (16.16,47.20)21.23 (11.04,44.62)0.13719.32(8.18,35.85)15.15 (6.03,38.59)0.633Hyperprolactinemia (n, %)28(63.6%)27(51.9%)0.24823(52.2%)19(36.5%)0.122Insufficient GH secretion (n, %)4(9.0%)6(11.5%)0.95513(29.5%)7(13.4%)0.053Hypogonadism (n, %)24(70.6%)[n = 34]23(51.1%)[n = 45]0.08133(97.1%)[n = 34]43(95.6%)[n = 45]1.000Hypothyroidism (n, %)12(27.2%)17(32.6%)0.56420(45.4%)31(59.6%)0.166Hypoadrenalism (n, %)13(29.5%)15(28.8%)0.94026(59.0%)30(57.6%)0.890Continuous variables with normal distribution were expressed as mean ± SD, non-normal data were expressed as median (interquartile range, IQR)BMI body mass index, FPG fasting blood glucose, HDL-c high-density lipoprotein cholesterol,LDL-c low-density lipoprotein cholesterol, TG triglycerides, TC total cholesterol, ALT alanine aminotransferase, AST aspartate aminotransferase, GGT gamma-glutamyl transferase, UA uric acid, ACTH adrenocorticotropic hormone, COR cortisol, FT3 free triiodothyronine, FT4 free thyroxine, TSH thyroid-stimulating hormone, GH growth hormone. Corrected Ca2+ = Total serum calcium + 0.02 × (40 − Serum albumin) Comparison of clinical characteristics of HO patients and non-HO patients in the training cohort Continuous variables with normal distribution were expressed as mean ± SD, non-normal data were expressed as median (interquartile range, IQR) BMI body mass index, FPG fasting blood glucose, HDL-c high-density lipoprotein cholesterol, LDL-c low-density lipoprotein cholesterol, TG triglycerides, TC total cholesterol, ALT alanine aminotransferase, AST aspartate aminotransferase, GGT gamma-glutamyl transferase, UA uric acid, ACTH adrenocorticotropic hormone, COR cortisol, FT3 free triiodothyronine, FT4 free thyroxine, TSH thyroid-stimulating hormone, GH growth hormone. Corrected Ca2+ = Total serum calcium + 0.02 × (40 − Serum albumin) Table 2Comparison of tumor characteristics of HO patients and non-HO patients in the training cohortVariablesHO group(n = 44)non-HO group(n = 52)P valueTumor type (n, %) Q S T9(20.5%)8(18.2%)27(61.4%)13(25.0%)18(34.6%)21(40.4%)0.0960.5980.0710.041*Necrosis (n, %)17(38.6%)26(50.0%)0.265Liquid level or sediment(n, %)6(13.6%)7(13.4%)0.980Tumor calcification (n, %)17(38.6%)15(28.8%)0.311Pituitary gland visualized(n, %)32(72.7%)37(71.1%)0.864Pituitary stalk visualized(n, %)9(20.4%)7(13.4%)0.360PSSP (n, %)29(65.9%)37(71.1%)0.581Intraoperative blood loss ≥ 100 ml(n, %)41(93.2%)35(67.3%)0.002*Tumor volume (cm3)14.54(6.22, 19.53)6.96(3.10, 19.32)0.040* Small tumor<5cm3 (n, %)2(4.5%)18(34.6%)< 0.001*Tumor texture (n, %)0.555 Cystic Cystic-Solid Solid16(36.4%)26(59.1%)2(4.5%)22(42.3%)28(53.8%)2(3.8%)0.5530.6061.000T1 (n, %)0.231 Isointense Hypointense Hyperintense27(61.4%)15(34.1%)2(4.5%)24(46.2%)25(48.1%)3(5.8%)0.1370.1661.000Extent of Resection (n, %)0.530 Partial Resection STR GTR3(6.8%)14(31.8%)27(61.4%)5(9.6%)11(21.1%)36(69.2%)0.9020.2350.419Grading ofsupratentorial hydrocephalus (n, %)0.959 Absent Mild Moderate Severe28(63.6%)7(15.9%)6(13.6%)3(6.8%)33(63.5%)8(15.4%)7(13.5%)4(7.7%)0.9860.9440.9801.000Preoperative Puget classification (n, %)0.035* Grade 0 Grade 1 Grade 23(6.8%)27(61.4%)14(31.8%)13(25%)28(53.8%)11(21.2%)0.017*0.4580.235Postoperative Puget classification (n, %)0.861 Grade 0 Grade 1 Grade 212(27.3%)22(50.0%)10(22.7%)14(26.9%)25(48.1%)13(25.0%)0.9690.8510.795Continuous variables with normal distribution were expressed as mean ± SD, non-normal data were expressed as median (interquartile range, IQR)Q-type infradiaphragmatic, S-type suprasellar-subarachnoid, T-type tuberoinfundibular, PSSP postoperative pituitary stalk preservation, T1 T1 signal characteristics of cystic fluid, STR Subtotal Resection, GTR gross total resection Table 2Comparison of tumor characteristics of HO patients and non-HO patients in the training cohortVariablesHO group(n = 44)non-HO group(n = 52)P valueTumor type (n, %) Q S T9(20.5%)8(18.2%)27(61.4%)13(25.0%)18(34.6%)21(40.4%)0.0960.5980.0710.041*Necrosis (n, %)17(38.6%)26(50.0%)0.265Liquid level or sediment(n, %)6(13.6%)7(13.4%)0.980Tumor calcification (n, %)17(38.6%)15(28.8%)0.311Pituitary gland visualized(n, %)32(72.7%)37(71.1%)0.864Pituitary stalk visualized(n, %)9(20.4%)7(13.4%)0.360PSSP (n, %)29(65.9%)37(71.1%)0.581Intraoperative blood loss ≥ 100 ml(n, %)41(93.2%)35(67.3%)0.002*Tumor volume (cm3)14.54(6.22, 19.53)6.96(3.10, 19.32)0.040* Small tumor<5cm3 (n, %)2(4.5%)18(34.6%)< 0.001*Tumor texture (n, %)0.555 Cystic Cystic-Solid Solid16(36.4%)26(59.1%)2(4.5%)22(42.3%)28(53.8%)2(3.8%)0.5530.6061.000T1 (n, %)0.231 Isointense Hypointense Hyperintense27(61.4%)15(34.1%)2(4.5%)24(46.2%)25(48.1%)3(5.8%)0.1370.1661.000Extent of Resection (n, %)0.530 Partial Resection STR GTR3(6.8%)14(31.8%)27(61.4%)5(9.6%)11(21.1%)36(69.2%)0.9020.2350.419Grading ofsupratentorial hydrocephalus (n, %)0.959 Absent Mild Moderate Severe28(63.6%)7(15.9%)6(13.6%)3(6.8%)33(63.5%)8(15.4%)7(13.5%)4(7.7%)0.9860.9440.9801.000Preoperative Puget classification (n, %)0.035* Grade 0 Grade 1 Grade 23(6.8%)27(61.4%)14(31.8%)13(25%)28(53.8%)11(21.2%)0.017*0.4580.235Postoperative Puget classification (n, %)0.861 Grade 0 Grade 1 Grade 212(27.3%)22(50.0%)10(22.7%)14(26.9%)25(48.1%)13(25.0%)0.9690.8510.795Continuous variables with normal distribution were expressed as mean ± SD, non-normal data were expressed as median (interquartile range, IQR)Q-type infradiaphragmatic, S-type suprasellar-subarachnoid, T-type tuberoinfundibular, PSSP postoperative pituitary stalk preservation, T1 T1 signal characteristics of cystic fluid, STR Subtotal Resection, GTR gross total resection Comparison of tumor characteristics of HO patients and non-HO patients in the training cohort Q S T 9(20.5%) 8(18.2%) 27(61.4%) 13(25.0%) 18(34.6%) 21(40.4%) 0.096 0.598 0.071 0.041* Liquid level or sediment (n, %) Pituitary gland visualized (n, %) Pituitary stalk visualized (n, %) Intraoperative blood loss ≥ 100 ml (n, %) Cystic Cystic-Solid Solid 16(36.4%) 26(59.1%) 2(4.5%) 22(42.3%) 28(53.8%) 2(3.8%) 0.553 0.606 1.000 Isointense Hypointense Hyperintense 27(61.4%) 15(34.1%) 2(4.5%) 24(46.2%) 25(48.1%) 3(5.8%) 0.137 0.166 1.000 Partial Resection STR GTR 3(6.8%) 14(31.8%) 27(61.4%) 5(9.6%) 11(21.1%) 36(69.2%) 0.902 0.235 0.419 Grading of supratentorial hydrocephalus (n, %) Absent Mild Moderate Severe 28(63.6%) 7(15.9%) 6(13.6%) 3(6.8%) 33(63.5%) 8(15.4%) 7(13.5%) 4(7.7%) 0.986 0.944 0.980 1.000 Grade 0 Grade 1 Grade 2 3(6.8%) 27(61.4%) 14(31.8%) 13(25%) 28(53.8%) 11(21.2%) 0.017* 0.458 0.235 Grade 0 Grade 1 Grade 2 12(27.3%) 22(50.0%) 10(22.7%) 14(26.9%) 25(48.1%) 13(25.0%) 0.969 0.851 0.795 Continuous variables with normal distribution were expressed as mean ± SD, non-normal data were expressed as median (interquartile range, IQR) Q-type infradiaphragmatic, S-type suprasellar-subarachnoid, T-type tuberoinfundibular, PSSP postoperative pituitary stalk preservation, T1 T1 signal characteristics of cystic fluid, STR Subtotal Resection, GTR gross total resection Baseline characteristics of the training and validation cohorts The clinical characteristics and tumor features of the training cohort are presented in Tables 1 and 2. Compared to the non-HO group, the HO group exhibited significantly higher preoperative BMI, serum sodium, uric acid levels, and postoperative serum calcium levels (all P < 0.05). Additionally, the HO group demonstrated a higher probability of preoperative hypothalamic involvement, intraoperative blood loss ≥ 100 mL, T-type tumors, and non-small tumors (all P < 0.05). No significant differences in gender distribution or age were observed between the groups (P > 0.05). Gonadal function was assessed in 79 cases (46 males, 33 females) within the training cohort, with no statistically significant differences in pre- or postoperative hypogonadism rates between the two groups (P > 0.05). In the validation cohort (Tables 3 and 4 in the appendix), most indicator changes remained consistent with those observed in the training cohort. However, unlike the training cohort, the validation cohort showed a significantly higher probability of absence of preoperative hydrocephalus in the non-HO group, with a statistically significant difference (P < 0.05). No statistically significant differences were observed between the two groups in preoperative serum sodium, preoperative serum uric acid, postoperative serum calcium levels, or the probability of T-type tumors. In the validation cohort, gonadal function was evaluated in 35 cases (19 males and 16 females), and there were no statistically significant differences in preoperative or postoperative hypogonadism between the two groups (P > 0.05). Table 3Univariate and multivariate logistic regression analysis of the postoperative HO groupVariablesUnivariate logistic regression analysisMultivariate logistic regression analysisOR (95%CI)P valueOR (95%CI)P valuePreoperative Na+1.069(0.961–1.190)0.220Preoperative BMI1.169(1.050–1.300)0.004*1.469(1.184–1.823)<0.001*Postoperative corrected Ca2+1.034(1.007–1.063)0.014*Preoperative UA1.006(1.001–1.012)0.019*Preoperative Puget gradegrade 00.220(0.058–0.830)0.025*T-type tumor2.345(1.031–5.332)0.042*Age of onset0.984(0.963–1.004)0.1200.922(0.879–0.966)0.001*Intraoperative blood loss ≥ 100 ml6.638(1.795–24.544)0.005*5.689(1.054–30.698)0.043*Small tumor0.090(0.019–0.415)0.002*0.122(0.018–0.838)0.032*OR Odds ratio, CI confident interval Table 3Univariate and multivariate logistic regression analysis of the postoperative HO groupVariablesUnivariate logistic regression analysisMultivariate logistic regression analysisOR (95%CI)P valueOR (95%CI)P valuePreoperative Na+1.069(0.961–1.190)0.220Preoperative BMI1.169(1.050–1.300)0.004*1.469(1.184–1.823)<0.001*Postoperative corrected Ca2+1.034(1.007–1.063)0.014*Preoperative UA1.006(1.001–1.012)0.019*Preoperative Puget gradegrade 00.220(0.058–0.830)0.025*T-type tumor2.345(1.031–5.332)0.042*Age of onset0.984(0.963–1.004)0.1200.922(0.879–0.966)0.001*Intraoperative blood loss ≥ 100 ml6.638(1.795–24.544)0.005*5.689(1.054–30.698)0.043*Small tumor0.090(0.019–0.415)0.002*0.122(0.018–0.838)0.032*OR Odds ratio, CI confident interval Univariate and multivariate logistic regression analysis of the postoperative HO group Preoperative Puget grade grade 0 OR Odds ratio, CI confident interval Table 4Age-stratified patient characteristicsVariables< 18 years (N = 30)≥ 18years (N = 106)Preoperative BMI (kg/m2) HO group18.28(16.35, 23.31)27.44(24.67, 28.89) non-HO group16.60(16.00, 18.32)23.43(21.83, 24.64) P value0.196< 0.001*Intraoperative blood loss≥100 ml (n, %) HO group15(78.9%)47(100%) non-HO group6(54.6%)41(69.49%) P value0.225< 0.001*Age of onset (years) HO group8.00 ± 3.8344.00(31.00, 55.50) non-HO group10.73 ± 3.9347.00(34.50, 57.00) P value0.0730.358Small tumor<5cm3 (n, %) HO group3(15.79%)2(4.26%) non-HO group3(27.27%)25(42.37%) P value0.641< 0.001* Table 4Age-stratified patient characteristicsVariables< 18 years (N = 30)≥ 18years (N = 106)Preoperative BMI (kg/m2) HO group18.28(16.35, 23.31)27.44(24.67, 28.89) non-HO group16.60(16.00, 18.32)23.43(21.83, 24.64) P value0.196< 0.001*Intraoperative blood loss≥100 ml (n, %) HO group15(78.9%)47(100%) non-HO group6(54.6%)41(69.49%) P value0.225< 0.001*Age of onset (years) HO group8.00 ± 3.8344.00(31.00, 55.50) non-HO group10.73 ± 3.9347.00(34.50, 57.00) P value0.0730.358Small tumor<5cm3 (n, %) HO group3(15.79%)2(4.26%) non-HO group3(27.27%)25(42.37%) P value0.641< 0.001* Age-stratified patient characteristics In our cohort of postoperative HO patients, hyperphagia (n = 87.8%) emerged as the most prevalent manifestation, followed by sleep disturbances predominantly manifesting as daytime hypersomnolence (n = 56.1%). Neuropsychiatric alterations including mood disorders (n = 50%) and cognitive-behavioral abnormalities (n = 37.8%) were also frequently observed. Development of the predictive model Univariate logistic regression analysis identified preoperative BMI, preoperative serum sodium, preoperative serum uric acid, postoperative serum calcium levels, T-type tumors, hypothalamic involvement, and intraoperative blood loss ≥ 100 ml as risk factors for HO, while small tumor was a protective factor (all P < 0.05). Following multicollinearity analysis, all factors exhibited variance inflation factors (VIF) < 10, confirming independence and absence of collinearity. Incorporating statistically significant indicators from the univariate logistic regression analysis and age at disease onset (previously validated in the literature) into the multivariate logistic regression model revealed four independent predictors with statistically significant differences: preoperative BMI, age at disease onset, intraoperative blood loss ≥ 100 ml, and small tumor size (P < 0.05) (Table 3). The predictive model for postoperative HO secondary to craniopharyngioma was established based on these four variables: preoperative BMI, age at disease onset, intraoperative blood loss (coded as 0 for < 100 ml and 1 for ≥ 100 ml), and tumor size (coded as 0 for non-small tumors and 1 for small tumors) (Fig. 2). Fig. 2Predictive model for postoperative HO. The points of each feature were added to obtain the total points, and a vertical line was drawn on the total points to obtain the corresponding risk of intraoperative blood loss, Preoperative BMI, Tumors volume, Age of onset Predictive model for postoperative HO. The points of each feature were added to obtain the total points, and a vertical line was drawn on the total points to obtain the corresponding risk of intraoperative blood loss, Preoperative BMI, Tumors volume, Age of onset Discriminative ability and calibration of the predictive model The area under the receiver operating characteristic curve (AUC) was calculated for both the training and validation cohorts. In the training cohort, the AUC was 0.865 (95%CI: 0.793–0.937) with a sensitivity of 79.5% and specificity of 80.8%, the internal validation using the Bootstrap method with 1000 resampling iterations demonstrated an AUC of 0.828, a sensitivity of 74.9%, and a specificity of 75.2%, indicating good predictive performance of the model (Fig. 3A, B). Fig. 3The receiver operating characteristic (ROC) curve and calibration plot. A. ROC curve for training cohort. B. ROC curve for internal validation with 1000 Bootstrap resamplings in training cohort. C. ROC curve for validation cohort. D. Calibration plot for training cohort. E. Calibration plot for validation cohort The receiver operating characteristic (ROC) curve and calibration plot. A. ROC curve for training cohort. B. ROC curve for internal validation with 1000 Bootstrap resamplings in training cohort. C. ROC curve for validation cohort. D. Calibration plot for training cohort. E. Calibration plot for validation cohort In the validation cohort, the area under the receiver operating characteristic curve (AUC) was 0.864 (95%CI: 0.744–0.983), with sensitivity and specificity of 90.9% and 72.2%, respectively (Fig. 3C). The calibration curves demonstrated close agreement between the predicted and observed probabilities of HO occurrence, indicating good consistency (Fig. 3D, E). The Hosmer-Lemeshow (H-L) test was applied to assess the model’s goodness-of-fit, yielding P-values of 0.96 for the training cohort and 0.64 for the validation cohort. Both cohorts had P-values > 0.05, suggesting that the model’s predicted values aligned well with the observed data. Decision curve analysis for predictive models The decision curve analysis (DCA) of the predictive model is presented in Fig. 4. The model’s curve is significantly distant from both the “None line” and “All line”, indicating its superior clinical utility. Fig. 4The decision curve analysis (DCA) of the prediction model demonstrates its clinical utility. The y-axis represents the net benefit. The red curve corresponds to the predictive model for postoperative HO, the gray curve illustrates the hypothetical scenario where all patients are assumed to develop HO, and the black curve denotes the counterfactual assumption of no patients developing HO The decision curve analysis (DCA) of the prediction model demonstrates its clinical utility. The y-axis represents the net benefit. The red curve corresponds to the predictive model for postoperative HO, the gray curve illustrates the hypothetical scenario where all patients are assumed to develop HO, and the black curve denotes the counterfactual assumption of no patients developing HO To gain a better usage in clinical practice, we also made an online website (https://hypothalamicobesity.shinyapps.io/mfuzz/). Age-stratified patient characteristics As presented in Table 4, age-stratified analysis in this study indicated that the postoperative HO incidence rate was 63.33% (19/30) in patients under 18 years and 44.34% (47/106) in patients aged 18 years or older. In the adult group (≥ 18 years), preoperative high BMI, intraoperative blood loss ≥ 100 ml, and non-small tumors were significantly associated with postoperative HO (all P < 0.001), which validated the stability of the prediction model. Notably, the incidence rate of intraoperative blood loss ≥ 100 ml in the adult HO group reached 100%, suggesting that it may serve as a “key indicator” for screening adult HO. In the pediatric and adolescent group (< 18 years), preoperative high BMI, intraoperative blood loss ≥ 100 ml proportion, and younger age of onset showed trends consistent with adults, but no statistical differences were found (all P > 0.05). This might be related to the smaller sample size (n = 30). Thus, further verification with a larger sample of minors is required in the future. Subgroup analysis by histological type In the ACP subgroup, the median age at onset was significantly lower in patients who developed HO (28 years; IQR: 10, 47.5) compared to those who did not (39 years; IQR: 24, 54.5; p = 0.03). In the PCP subgroup, the mean age at onset was lower in the HO group (39.91 ± 18.15 years) than in the non-HO group (42.26 ± 18.19 years), although this difference was not statistically significant (p = 0.63). Based on the above age-stratified analysis results, we constructed a clinical management pathway for hypothalamic obesity after craniopharyngioma surgery (Fig. 5) in this study. The pathway uses the age of 18 years as a conventional decision-making boundary and outlines postoperative monitoring priorities based on the identified risk profiles of different age groups. In the group under 18 years of age, the focus is on the high-risk population aged ≤ 10 years (73.7% of HO children in this study were aged ≤ 10 years), who should receive intensive monitoring (monthly BMI measurement, sleep quality assessment, etc.). In the group aged 18 years and above, the pathway focuses on intraoperative blood loss control and preoperative BMI management. This pathway provides a conceptual and visual synthesis of our key findings, offering a framework to guide initial clinical attention. Its primary utility lies in identifying high-risk patients who may benefit from proactive intervention, which addresses the most immediate clinical need. This targeted approach aims to improve clinical efficiency by optimizing resource allocation. Building on this foundation, future studies should focus on validating and refining the pathway, including the formal definition of a low-risk stratum to further enhance its precision and applicability across the entire risk spectrum. Fig. 5Clinical management pathway for hypothalamic obesity after craniopharyngioma surgery Clinical management pathway for hypothalamic obesity after craniopharyngioma surgery Discussion HO is a common postoperative complication of craniopharyngioma that contributes to both metabolic disorders and psychosocial impairment. In this study, patients aged over 18 years accounted for 77.95% of the cohort, with a mean follow-up period of 33 months (range: 2–67 months). The overall incidence rate of HO was 48.53%, demonstrating a male-to-female ratio of 1.2:1, with no statistically significant difference observed. This study included 78 ACP cases and 58 PCP cases. We find that the prevalence of the papillary subtype in our series is higher than typical. This difference likely reflects two factors related to our study. First, as a large regional medical center, our single-institution cohort may not be fully representative of the general population, introducing a potential selection bias. Second, our study primarily comprised adult patients, a demographic in which a higher proportion of the papillary subtype is consistent with previous reports [22–24]. Previous studies have indicated that HO commonly develops within 6–12 months postoperatively, and is usually followed by a plateau phase [25–26]. In this study, the median time to postoperative onset of HO was 8.5 (5.75−12) months. The findings of this study revealed that older age at onset serves as a protective factor against postoperative HO in craniopharyngioma patients. Research has demonstrated that pediatric craniopharyngioma patients exhibit a significantly higher incidence rate of postoperative HO compared to adults, with those presenting an initial disease onset age < 10 years showing a substantially increased risk of developing obesity, visual impairment, and hypopituitarism following surgery [27]. According to the study by Guo et al. involving 335 surgically treated craniopharyngioma patients, under uniform surgical protocols, the incidence of postoperative HO reached 87.5% in pediatric cases, demonstrating a statistically significant elevation compared to the 55.5% prevalence observed in adult cohorts [28]. The developing brains of children undergo rapid maturation phases during which hypothalamic functions remain relatively immature, exhibiting reduced capacities for neural repair and homeostatic adaptation following injuries [29–30]. In addition, anatomical considerations are crucial; a tumor of a given absolute volume represents a greater relative burden within a smaller pediatric cranium, potentially causing more severe hypothalamic compression [1]. Consequently, even with comparable lesion severity, pediatric patients are more vulnerable to developing hypothalamic dysfunction and consequent obesity compared to adults. In this study, the incidence rate among adults was 44.33%, while children and adolescents under 18 years old and children under 10 years old showed significantly higher incidence rates of 63.33% and 72.22% respectively. These findings are consistent with previous research. Univariate regression analysis in this study demonstrated no significant age disparity between groups. However, when incorporating age at disease onset (previously validated in the literature) into multivariate regression analysis, younger age at onset persisted as an independent risk factor for postoperative HO. This paradoxical association may reflect residual confounding from unmeasured variables. To investigate whether the predictive value of “age at onset” is independent of “histological subtype,” we performed a subgroup analysis. Within both the ACP and PCP subgroups, a younger age at onset continued to show an association with a higher risk of HO. This consistent trend suggests that younger age itself may remain a risk factor even within the same histological subtype. In summary, we posit that age at onset and histological type are interrelated yet distinct predictive factors. Age at onset may represent an independent, physiological susceptibility factor. Consequently, patients with a younger age at disease onset warrant vigilant monitoring of anthropometric parameters, including weight trajectory and appetite fluctuations, to facilitate timely nutritional interventions that mitigate excessive weight gain. Electrolyte disturbance is one of the most common postoperative complications in patients with craniopharyngioma. The urine volume and electrolyte level should be routinely and dynamically monitored after surgery, and corrected in time according to the type of electrolyte disorder. While previous research has predominantly concentrated on serum sodium (Na+) and serum chloride (Cl−) imbalances, there remains a paucity of studies investigating postoperative serum calcium (Ca2+) and serum magnesium (Mg2+) electrolyte disturbances in craniopharyngioma patients. In this study, postoperative day 1 measurements revealed elevated Na + and Cl− levels above preoperative baselines, while Ca2+, K+, and Mg2+ concentrations demonstrated reductions below baseline measurements. This electrolyte pattern is hypothesized to be associated with the early postoperative elevation in serum sodium levels. Postoperative patients in the HO group demonstrated higher serum calcium levels, potentially attributable to their elevated albumin levels. Our findings demonstrate that preoperative elevated BMI constitutes an independent risk factor for postoperative HO in craniopharyngioma patients. This observation aligns with established evidence in the literature regarding metabolic predictors of HO development [7, 31–32]. Accordingly, we propose an integrated “two-hit” model to explain the link between preoperative elevated BMI and postoperative HO. In this model, the first hit comprises both the direct hypothalamic compression by the tumor and a pre-existing state of metabolic dysregulation often reflected by high BMI. This metabolic state, characterized by adipose-derived pro-inflammatory cytokines (e.g., TNF-α, IL−6) that can cross the blood-brain barrier (BBB) and impair hypothalamic nuclei, establishes a “vulnerable hypothalamus” [33–36]. Consequently, some patients may present with obesity at diagnosis, which serves as a surrogate indicator of this preclinical hypothalamic dysfunction. The second hit is the surgical intervention itself, which inflicts unavoidable injury. This trauma directly disrupts appetite-regulating nuclei (e.g., arcuate nucleus), triggering primary hyperphagia and rapid weight gain. The resultant weight gain promotes systemic insulin resistance, which, through dysregulation of central insulin signaling, may further exacerbate hyperphagia, creating a vicious metabolic cycle that culminates in overt HO [37–38]. Notably, the mass effect and associated inflammation of craniopharyngiomas can lead to elevated BMI in a proportion of patients even before surgical intervention. Previous studies have established that obesity identified at diagnosis is a significant risk factor for HO and serves as a surrogate indicator of hypothalamic dysfunction caused by the tumor mass effect [5–6]. Preoperative BMI indirectly reflects the extent of hypothalamic involvement by the tumor. Clinical studies demonstrate that 12%−19% of patients are classified as obese at diagnosis, with this proportion escalating to 55% following surgical intervention [39–40]. Our study revealed that preoperative obese patients experienced rapid weight gain (ΔBMI > 10%) during the early postoperative period, accompanied by emerging non-endocrine hypothalamic dysfunction. This suggests that postoperative weight gain may not merely represent a continuation of preoperative obesity, but rather results from further weight gain induced by postoperative hypothalamic dysfunction. Previous studies have reported that postoperative HO typically occurs 6–12 months after surgery. However, to account for individual variability and potential delayed onset, this study defines HO as weight gain meeting obesity criteria within 1 year following tumor resection, thereby comprehensively capturing weight changes during this timeframe. Therefore, by integrating quantifiable weight changes with clinical symptoms, this composite definition of HO allowed for a more comprehensive and reliable evaluation of its development after surgery. The key to successful resection of craniopharyngioma lies in the precise identification and delineation of the tumor’s relationship with the hypothalamic-pituitary stalk and hypothalamic tissue. Typically, complete and safe resection can only be achieved through sharp dissection under direct visualization. Sharp dissection of the tumor from the surrounding brain tissue inevitably damages neurovascular tissues, compromising patient prognosis. Previous studies have demonstrated that intraoperative hemorrhage is one of the most significant visual obstacles during surgery [41]. Many skull base tumors possess a robust vascular supply, which contributes significantly to blood loss and compromised surgical visibility, impairing the identification of local anatomical landmarks and increasing the risk of intraoperative hypothalamic injury [42]. This study identified intraoperative blood loss ≥ 100 ml as another independent risk factor for postoperative HO in craniopharyngioma. This threshold was selected based on clinical practicality, as it represents a round figure that facilitates intraoperative risk stratification in neurosurgical practice. Notably, this threshold demonstrated a significant association with HO risk in our dataset (P < 0.05), supporting its clinical utility. Excessive intraoperative hemorrhage may lead to impaired hypothalamic functional recovery and elevate the risk of hypothalamic obesity. Significant intraoperative bleeding could induce postoperative hypoxia and ischemia in brain tissue, further impairing the hypothalamic energy homeostasis regulatory function [43–44]. Substantial intraoperative blood loss may reflect greater surgical complexity, underscoring the need for vigilant postoperative metabolic monitoring and early mobilization to mitigate weight gain risk [45–47]. Therefore, comprehensive postoperative management is essential. For patients identified as high-risk by our model, this includes consideration of hypothalamus-sparing strategies—such as planned subtotal resection with adjuvant radiotherapy—along with meticulous surgical hemostasis to minimize hypothalamic injury, followed by close postoperative monitoring of weight trajectory and appetite to enable early intervention. Studies have demonstrated that the tumor location and size in patients with craniopharyngioma are correlated with the severity of hypothalamic obesity. This research applied a novel classification system, namely the QST classification [48]. This system was established based on the biological characteristics of craniopharyngioma, including its origin, growth patterns, and relationships with adjacent anatomical structures, particularly focusing on tumor origination from membranous structures related to the sellar region. The QST classification holds significant clinical value for guiding surgical strategy selection, prognostic evaluation, and analysis of endocrine characteristics in craniopharyngioma patients. T-type craniopharyngiomas frequently occupy the third ventricular cavity, leading to a higher severity of postoperative HO in these patients. In contrast, Q-type and S-type craniopharyngiomas exhibit less hypothalamic involvement, resulting in a relatively lower incidence of severe obesity [49]. In this study, T-type tumor was identified as a risk factor for postoperative HO; however, it did not emerge as an independent risk factor upon multivariate regression analysis and was thus excluded from the final regression model. This finding should not be interpreted as a diminished clinical relevance of the T-type location, the risk conveyed by T-type tumors may be largely captured by more precise and direct metrics. This phenomenon is consistent with that reported in the larger study by Qi’s team, whose final predictive model also selected direct quantitative measures over tumor subtype [7]. However, consideration should be given to the fact that our cohort size is small, the statistical power is limited. The contribution of T-type tumors to HO risk requires further validation in larger, prospective cohorts. Regarding tumor volume, this study found that tumor volume < 5 cm³ serves as a protective factor against postoperative HO in craniopharyngioma patients. While tumor volume is a continuous variable, we chose to dichotomize it to improve the clinical interpretability and utility of our predictive model, allowing for straightforward risk stratification (“small” vs. “non-small”). The threshold of 5 cm³ was selected based on its clinical relevance and the general association of tumors below this volume with a higher likelihood of achieving gross-total resection, thereby reducing the risk of hypothalamic injury. This round figure is also easily communicable and routinely used by neurosurgeons for preliminary assessment. Patients with smaller tumor volumes demonstrated better postoperative hypothalamic functional recovery and a lower risk of developing HO. Previous studies have shown that larger tumor volumes are more likely to cause hypothalamic dysfunction and poorer prognosis. This may be attributed to mechanical compression of the hypothalamus by larger tumors, increased intrasellar pressure compressing portal vessels, and subsequent reduction in blood supply to normal pituitary tissue [50–52]. Additionally, smaller tumor volumes are associated with reduced surgical difficulty and minimized intraoperative hypothalamic injury. Patients with larger tumors experience significant weight gain during follow-up compared to preoperative baselines, and prolonged follow-up may reveal progression to HO. This phenomenon may be attributed to the increased likelihood of hypothalamic involvement in cases with larger tumor volumes. In the present study, postoperative Puget classification did not emerge as an independent predictor in the final multivariate model. A potential explanation is that the Puget classification only grossly describes the degree of hypothalamic injury and lacks direct correlation with specific hypothalamic nuclei. Therefore, its isolated use for predicting postoperative HO is not advisable. Notably, similar findings have been reported in previous studies, in which Puget classification also did not retain independent predictive value in the final multivariable model [32]. This suggests that in multivariable modeling, the predictive contribution of Puget classification may be partially substituted when more direct or quantifiable measures are incorporated. In terms of the clinical characteristics of the training and validation cohorts in this study, as shown in Appendix Tables 1 and 2, most clinical and tumor features were generally well balanced between the two cohorts, and statistically significant differences were observed in a limited number of baseline variables. These differences likely reflect random variation due to the smaller size of the validation cohort (n = 40) and the natural differences that exist among patients in a single-center, retrospective series. Importantly, none of these imbalanced variables were included as predictors in the final multivariate model, suggesting that their between-cohort differences did not substantially bias model development or validation. The overall comparability of the two cohorts supports the reliability of the internal validation results. In the era of precision medicine, integrating our predictive model with the context of emerging targeted therapies for craniopharyngioma is crucial. It is well-established that nearly all PCPs involve dysregulated BRAF-V600E signaling, whereas ACPs are typically characterized by the disordered Wnt/β-catenin signaling pathway [53]. For PCPs, combination therapy with a BRAF inhibitor (e.g., dabrafenib) and a MEK inhibitor (e.g., trametinib) has demonstrated remarkable efficacy in recurrent or advanced cases. For eligible patients with PCP, neoadjuvant targeted therapy can significantly reduce tumor volume and lessen hypothalamic adhesion, thereby facilitating safer subsequent surgery with a higher likelihood of gross total resection and maximal hypothalamic preservation [54–55]. While direct Wnt/β-catenin targeting in ACPs is still being explored, therapeutic strategies have shifted toward modulating its prominent inflammatory tumor microenvironment. Agents such as tocilizumab (an IL−6 receptor inhibitor) and bevacizumab (an anti-angiogenic agent) represent promising exploratory directions [56]. Therefore, for patients identified as high-risk by our model, a multidisciplinary team evaluation should assess the feasibility of neoadjuvant targeted therapy, enabling a personalized strategy that balances tumor control with hypothalamic preservation. Existing prediction models predominantly focus on a limited range of factors. They may not fully capture the significant influence of preoperative patient metabolic status or critical intraoperative details on postoperative HO development, potentially limiting prediction comprehensiveness. Furthermore, reliance on complex or difficult-to-acquire variables hinders clinical adoption and reduces model utility [7, 32, 57]. The prediction model developed in this study, incorporating preoperative BMI, age at onset, intraoperative blood loss, and tumor volume, demonstrated excellent predictive performance (AUC = 0.865, sensitivity = 79.5%, specificity = 80.8%). In contrast to qualitative classifications that rely on postoperative imaging (e.g., Puget grade), our model integrates readily obtainable preoperative, intraoperative, and tumor characteristics to identify high-risk patients in clinical practice who may benefit from closer monitoring, earlier nutritional intervention, or hypothalamus-sparing surgical strategies. Deployment as a web-based calculator significantly enhances ease of use and facilitates practical implementation. A limitation is the model’s limited capacity for direct hypothalamic assessment, including quantitative MRI evaluation. Future iterations could be enhanced by incorporating radiomics parameters. Additionally, the growing application of emerging technologies, particularly artificial intelligence (AI), offers significant opportunities in medical technology. Incorporating AI-driven large-scale data analysis and deep learning algorithms could substantially improve the predictive accuracy and reliability of models for hypothalamic obesity. The strength of this study lies in the development of a risk stratification system for early identification of postoperative HO in craniopharyngioma patients. By incorporating both preoperative and postoperative clinical parameters, our model demonstrates greater comprehensiveness compared to previous studies. Notably, intraoperative blood loss ≥ 100 mL was identified as a significant risk factor for HO. Furthermore, the clinically accessible parameters included in this model enhance its feasibility for widespread clinical adoption. However, this study has several limitations. First, growth hormone deficiency (GHD) was defined as a peak GH level < 10ng/mL in children/adolescents undergoing GH stimulation tests and <5ng/mL in adults [58]. Due to the absence of GH stimulation testing and substantial missing data on insulin-like growth factor-1 (IGF-1) levels, IGF-1 was not incorporated into the current study design, and thus the precise GHD status within our cohort could not be conclusively determined. Based on previous studies and our institutional reference ranges for GH measurement (chemiluminescence immunoassay: normal range 0.06-5.0 ng/mL), we operationally defined insufficient GH secretion as levels < 0.06 ng/mL. Although no significant correlation was observed between this criterion and postoperative HO, this conclusion remains limited by methodological constraints, and its clinical relevance warrants further investigation. In subsequent phases, we will prospectively collect additional cases to validate IGF-1 profiles and further investigate the mechanistic role of GHD in HO pathogenesis. Second, the short postoperative follow-up duration in some patients, along with our use of complete-case analysis for missing data, may introduce bias—particularly if missing data are not completely at random. Third, as a single-center study with a relatively small sample size and lacking validation through external independent cohorts, the generalizability of our findings requires confirmation in broader analyses. While the sample size of this study is relatively modest, we took methodological precautions to ensure model stability. The events per variable (EPV) for our final model was 11 (44 events for 4 variables), a value that falls within the acceptable range (≥ 10) and helps to reduce concerns regarding overfitting. Although internal validation demonstrated promising model performance, external validation in independent, large-scale, multicenter cohorts is essential before clinical implementation—a step we plan to undertake in future prospective studies. Separately, interventional studies are also needed to determine whether comprehensive perioperative management, including intraoperative surgical optimization and early postoperative metabolic support, can reduce the risk of hypothalamic obesity in high-risk patients identified by our model. Conclusions In summary, this study identified preoperative BMI, age at onset, intraoperative blood loss, and tumor volume as clinically relevant predictors of postoperative HO in craniopharyngioma patients. Higher preoperative BMI and greater intraoperative blood loss ≥ 100 mL demonstrated significant associations with increased risk of postoperative HO in craniopharyngioma patients, whereas advanced age at onset and smaller tumor volume (< 5 cm³) were correlated with reduced HO risk. The risk prediction model established in this study offers a preliminary tool to aid in identifying high-risk populations. Its utility in guiding targeted interventions for mitigating metabolic complications and optimizing long-term outcomes awaits confirmation in future studies. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1