We performed a retrospective observational study at a third level public hospital. The study window was January 2023 to December 2024. The institutional review board approved the protocol. Because of the retrospective design, informed consent was waived. All data were anonymized before analysis. The study reflected routine clinical practice and did not alter patient management.

Eligible patients had a colostomy suring Hartmann procedure with adequate conditions for reconnection and complete clinical and biochemical information. We included consecutive cases within the study window. Patients were identified through operative logs and electronic records. Exclusion criteria were incomplete medical CRP values on both postoperative day 1 (POD1) and postoperative day 3 (POD3), active malignancy, chronic immunosuppressive therapy (including long-term systemic corticosteroids or other immunomodulators), or a concurrent infectious process unrelated to surgery. These criteria were chosen to reduce baseline inflammatory noise and isolate surgery-related changes in CRP; the final cohort represents all cases that met these criteria during the period.

The main exposure was the change in CRP between POD1 and POD3. We expressed this as ΔCRP, defined as the absolute difference between CRP values measured on postoperative day (POD) 1 and POD3, and we included only patients with complete records containing both measurements, which are routinely obtained at our institution. Sampling times were not protocolized to a fixed clock time and corresponded to the CRP determinations documented for each respective postoperative day in routine care; because all assessed patients fulfilled the completeness criterion, no imputation for ΔCRP was required. Surgical success was defined a priori as an uncomplicated postoperative course, operationalized as absence of major complications, no unplanned reoperation within 30 days, and discharge within the expected postoperative length of stay for Hartmann reversal (≤7 postoperative days at our institution). Major complications were defined as postoperative events requiring surgical, endoscopic, or radiologic intervention, admission to the intensive care unit, or resulting in organ failure or death; prespecified events included anastomotic leak, prolonged ileus requiring nasogastric decompression beyond postoperative day 5, intra-abdominal abscess, and deep or organ-space surgical site infection. We did not formally grade complications using the Clavien–Dindo classification, but used this pragmatic definition based on intervention and level of care, which reflects routine decision-making after Hartmann reversal. Surgical failure (SF) was defined as the presence of any complication during that period. These definitions align with routine decision making after HR.

We abstracted age, sex, and body mass index from the electronic medical record. Preoperative laboratory variables included hemoglobin, hematocrit, leukocytes, platelets, and albumin. Surgical variables included indication for HR, operative approach, duration of surgery, and estimated blood loss. Postoperative CRP was recorded on POD1 and POD3. Data abstraction followed a standardized template to maintain consistency filled by an independent blinded team. When a data field was unclear, the source note in the record was reviewed to confirm the value.

This observational cohort included all eligible cases in the study period. With 83 patients and ∼30% SF, there were approximately 25 events, permitting adequate logistic regression. post hoc precision analysis indicates that, for an area under the receiver operating characteristic (ROC) curve (AUC) near 0.75, adequate to judge discriminative value and motivate prospective validation.

We also used ROC analysis to estimate the discriminative performance of ΔCRP and to identify an optimal cutoff based on Youden's Index. To assess internal validity, we performed bootstrap resampling with 2,000 iterations, recalculating the AUC and Youden's optimal cut-off in each resample. We summarized the bootstrap distribution using mean and percentile-based 95% confidence intervals (CI). In addition, we conducted a sensitivity analysis by evaluating the performance of nearby ΔCRP thresholds (±5%–10% around the optimal cut-off), reporting sensitivity, specificity, PPV, NPV and confusion-matrix counts for each threshold.

Continuous variables were summarized as means with standard deviations or as medians with interquartile ranges, according to distribution. Categorical variables were reported as counts and percentages. Comparisons between surgical success and surgical failure used Student t-test or Mann–Whitney U test for continuous data and chi square or Fisher exact test for categorical data. We fitted multivariable logistic regression models to evaluate the association between ΔCRP and surgical success. Candidate covariates included body mass index (BMI), surgical time, intraoperative bleeding, serum albumin, neutrophil-to-lymphocyte ratio (NLR), and platelet-to-lymphocyte ratio (PLR). We examined multicollinearity using variance inflation factors (VIFs). Model calibration and overall fit were evaluated using the Hosmer–Lemeshow goodness-of-fit test and pseudo-R² indices (McFadden, Cox–Snell, and Nagelkerke). We inspected standardized residuals and influence statistics to assess major violations of logistic regression assumptions. Statistical significance was set at p < 0.05.

Analyses and plots were performed using IBM SPSS Statistics for Windows version 26 (Chicago IL. USA), RStudio 2023.03.0 + 386 “Cherry Blossom” for Windows.

We analyzed 83 patients who underwent Hartmann reversal during the study window. Among the 25 patients (30%) experienced a postoperative complication. Reintervention was the most frequent event (n = 13; 52%). Anastomotic leakage occurred in (n = 9; 36%). Additionally, 96% of patients with SF experienced other postoperative adverse events (Supplementary Table S1).

Baseline characteristics were comparable between groups. Median age was 49 years (IQR: 41–60) and 57% were male. Sex and age distribution did not differ between surgical success and surgical failure. Body mass index showed a similar pattern across groups. The main original indications for colostomy were complicated diverticular disease in 33 patients (40%), tumor in 6 patients (7.2%), and ballistic trauma in 8 patients (9.6%). There were no significant differences in indication profiles between outcomes (Table 1).

Laparoscopic HR was performed in 52 patients (63%). Open reversal was performed in 26 patients (31%). The rest of the reported cases started as minimal invasion but were converted to an open surgical approach and conversion occurred in 6%. Surgical approach did not differ between groups (p = 0.766). The median surgical time was 184 min (IQR: 160–240; p = 0.073). Estimated blood loss was similar between groups. (p = 0.141). However, postoperative hospital stay was significantly longer in the patients with SF (8 days, IQR: 7–11) compared to the patients with SS (6 days, IQR: 5–8), (p = 0.001) (Table 2).

In univariable analysis, higher ΔCRP [0.989; (95% CI: 0. 982–0.995); p ≤ 0.001] and lower PLR [0.987; (95% CI: 0.974–0.996); p = 0.019] were associated with a lower probability of surgical success, whereas BMI, surgical time, intraoperative bleeding, albumin, and NLR were not significantly associated (Table 3). In the multivariable model including ΔCRP, NLR, and PLR, ΔCRP remained independently associated with surgical success (OR: 0.987; 95% CI 0.977–0.995; p = 0.003), while NLR and PLR were no longer statistically significant (Table 3). VIF values for all covariates were below 2, indicating no evidence of problematic multicollinearity (Supplementary Table S3) The final multivariable logistic regression model showed acceptable calibration (Hosmer–Lemeshow p = 0.093) and good overall fit, with McFadden, Cox–Snell, and Nagelkerke pseudo-R² values of 0.41, 0.41, and 0.57, respectively.

We evaluated the ability of ΔCRP to predict surgery success using ROC analysis. In Figure 2, the AUC is 0.754 (95% CI: 0.634–0.873), which reflects moderate discrimination. The SE is 0.061, and the AUC differs from 0.5 on the DeLong test with p = 3.07 × 10⁻⁵. The Youden rule selected a threshold of 113.1, giving sensitivity 84.5% and specificity 60.0% with a 40.0% false-positive rate. The Closest Top Left criterion suggested 84.0, yielding sensitivity 70.7% and specificity 72.0% with a 28.0% false-positive rate. At the Youden threshold, PPV was 83.1% and NPV was 62.5% in a sample where success prevalence was 69.9%. The corresponding likelihood ratios were 2.11 for LR+ and 0.26 for LR−, indicating that higher ΔCRP values increase the probability of success while lower values decrease it, and that results should be interpreted with clinical context.

In bootstrap internal validation, the mean AUC was 0.755 (95% CI of 0.631–0.867), consistent with the reported AUC. The mean optimal ΔCRP cut-off across bootstrap samples was 106.9 mg/L (95% CI: 48.5–169.6), indicating that the discriminatory region lies approximately within the 100–120 mg/L range rather than at a single exact value. In a sensitivity analysis of thresholds around the apparent Youden cut-off (113.1 mg/L), sensitivity ranged from 0.76 to 0.88 and specificity from 0.52 to 0.64, while PPV remained stable at ∼0.81–0.83 and NPV at ∼0.53–0.65, supporting the robustness of the model's performance in this range.

This study reflects real practice in a public tertiary center where Hartmann reversal is performed regularly. The cohort included consecutive patients within a defined window, which reduces selection bias and mirrors the case-mix seen by multidisciplinary teams. The exposure of interest was simple and available in routine care. CRP was measured on postoperative day 1 and postoperative day 3, which are standard points in many pathways and capture the early inflammatory arc. Using the change between these two time points provides an interpretable signal that clinicians can apply at the bedside without extra cost or specialized assays. The outcome definition focused on events that matter to patients and services. Absence of major complications, no reoperation, and discharge within the expected recovery period are tangible targets for teams planning postoperative care. The analysis used familiar summary measures and widely understood tests, which supports reproducibility and clinical uptake. Taken together, these features increase feasibility and support near term translation of the findings into decision support after HR.

Our single-center retrospective design and relatively small sample size (n = 83) limit causal inference, increase the risk of residual confounding, and may restrict generalizability because thresholds for intervention, postoperative surveillance, and discharge can differ across institutions. The sample size was modest, which affects precision of estimates and the stability of subgroup effects. The ΔCRP cut-off was derived and internally explored using bootstrap resampling and sensitivity analyses, which suggest a robust discriminatory region within the 100–120 mg/L range. However, these procedures were performed in a single-center retrospective cohort, and external validation in independent populations is still required before adopting specific decision thresholds in routine practice. Although we excluded patients with overt infection unrelated to surgery and those on chronic immunosuppressive therapy, possibly adding selection bias, we did not systematically capture minor preoperative infections or short-term perioperative steroid use, which may still influence CRP kinetics and contribute to unmeasured confounding. This raises the risk of overfitting and underscores the need for external validation in diverse settings. Timing of blood draws, perioperative fluids, analgesic regimens, and antibiotics can influence CRP values. Although measurements were obtained on postoperative day 1 and postoperative day 3, variation in the exact clock time may still introduce measurement bias and noise into ΔCRP estimates. Only CRP was evaluated. Other inflammatory markers and composite clinical scores were not assessed, so potential incremental value remains unknown. Long term outcomes were not captured, and the definition of surgical success included discharge within a five day period. Discharge practices and bed availability may influence this component and could vary by center or season. Finally, exclusion of cases with missing CRP on both time points may introduce selection bias toward patients with more complete monitoring. These constraints should temper interpretation and guide the design of prospective multicenter studies that test predefined thresholds and assess clinical impact.

ΔCRP is practical for routine use after HR because CRP is already drawn on postoperative day 1 and day 3 in many pathways. Calculating the difference and recording it in the daily note can support bedside judgment without new resources. A small or falling ΔCRP aligns with an uncomplicated course and can reinforce standard mobilization, diet advancement, and discharge planning within ERAS protocols (8, 21). A large or rising ΔCRP should prompt caution with closer observation, early imaging when indicated, and a focused search for infection or leak (16–18, 22). Lines and drains should be reviewed, fluid balance reassessed, and antibiotics verified. The threshold observed in this cohort was about 113 mg/L and should be used as a guide rather than a rule, interpreted with the clinical picture and local capacity (23). Teams can integrate ΔCRP into handovers and track outcomes over time to refine local cut points and workflows. The aim is earlier recognition of deterioration and safer, timely discharge after HR.

Future research should focus on validating the ΔCRP threshold in prospective, multicenter trials to enhance its generalizability and clinical applicability. Additionally, integrating CRP kinetics into machine learning-based risk prediction models could improve real-time patient monitoring and early complication detection. Further studies should also assess the utility of additional inflammatory biomarkers, such as procalcitonin or interleukin-6, to determine whether a multimarker approach enhances predictive accuracy. Moreover, investigating the impact of perioperative interventions, including ERAS protocols and immunonutrition, on CRP kinetics and surgical outcomes may provide insights into modifiable factors that could improve postoperative recovery and reduce complication rates.