Classification of postoperative fever patients in the intensive care unit following intra-abdominal surgery: a machine learning-based cluster analysis using the Medical Information Mart for Intensive Care (MIMIC)-IV database, developed in the United States

Sang Mok Lee; Hongjin Shim

doi:10.4266/acc.004464

Articles

Page Path: HOME > Acute Crit Care > Forthcoming articles > Article

Original Article Surgery Classification of postoperative fever patients in the intensive care unit following intra-abdominal surgery: a machine learning-based cluster analysis using the Medical Information Mart for Intensive Care (MIMIC)-IV database, developed in the United States: Sang Mok Lee^1,2, Hongjin Shim^1,2; DOI: https://doi.org/10.4266/acc.004464
Published online: April 30, 2025

¹Department of Acute Care Surgery, Korea University Guro Hospital, Seoul, Korea

²Department of Trauma Surgery, Armed Forces Trauma Center, Armed Forces Capital Hospital, Seongnam, Korea

Corresponding author: Hongjin Shim Department of Acute Care Surgery, Korea University Guro Hospital, 148 Gurodong-ro, Guro-gu, Seoul 08308, Korea Tel: +82-2-2626-1147, Fax: +82-2-2626-1148, E-mail: simong3@yonsei.ac.kr

• Received: November 29, 2024 • Revised: January 27, 2025 • Accepted: February 6, 2025

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

443 Views
4 Download

Full Article

Download PDF

Abstract
INTRODUCTION
MATERIALS AND METHODS
RESULTS
DISCUSSION
KEY MESSAGES
NOTES
REFERENCES

Abstract

Background
Postoperative fever is common. However, it can sometimes indicate severe complications such as sepsis or pneumonia. Intensive care unit (ICU) patients who have undergone abdominal surgery have a higher risk of postoperative fever due the physical severity of this type of surgery. Nevertheless, determining when more aggressive or invasive management of fever is necessary remains a challenge.
Methods
We analyzed the Medical Information Mart for Intensive Care (MIMIC)-IV and MIMIC-IV-Note databases, which are open critical care big databases from a single institute in the United States. From this, we selected ICU patients who developed fever after intra-abdominal surgery and classified these patients into two groups using cluster analysis based on diverse variables from the MIMIC-IV databases. Following this cluster analysis, we assessed differences among the identified groups.
Results
Of 2,858 ICU stays after intra-abdominal surgery, 331 postoperative fever cases were identified. These patients were clustered into two groups. Group A included older patients with a higher mortality rate, while group B consisted of younger patients with a lower mortality rate.
Conclusions
Postoperative ICU patients with a fever could be classified into two distinct groups, a high-risk group and low-risk group. The high-risk patient group was characterized by older age, higher Sequential Organ Failure Assessment (SOFA) score, and more unstable hemodynamic status, indicating the need for aggressive management. Clustering postoperative fever patients by clinical variables can support medical decision-making and targeted treatment to improve patient outcomes.
Key Words: cluster analysis; critical care; fever; postoperative care; surgery

INTRODUCTION

Postoperative fever is a common symptom in patients following surgery, with a wide range of causes and clinical courses. Although it often resolves without specific treatment, it also can be a precursor to serious complications such as sepsis and pneumonia [1,2]. In particular, patients admitted to the intensive care unit (ICU) after surgery require especially careful monitoring due to the physiological burden of the surgical procedure and the severity of the underlying medical conditions. However, because fever is a non-specific symptom, determining if it is necessary to perform intensive examinations and invasive treatments for the fever remains challenging. In this study, we analyzed the characteristics of patients admitted to the ICU after abdominal surgery using the Medical Information Mart for Intensive Care (MIMIC)-IV database [3-5] and MIMIC-IV-Note database [5,6]. MIMIC-IV and MIMIC-IV-Note are large, publicly available critical care databases of a single institute developed in the United States, containing detailed information about ICU patients including sex, age, vital signs, laboratory results, procedure records, radiologic findings, and discharge notes, making these databases particularly suitable for the analysis of postoperative outcomes and identification of risk factors. Using these two databases, we classified postoperative abdominal surgery patients who developed fever into two groups using a Gaussian Mixture Model (GMM), a machine learning-based classification technique. Through this analysis, we aimed to classify patients into groups and determine the distinct characteristics of these groups to facilitate the identification of patients who require more active and aggressive examinations or interventions.

MATERIALS AND METHODS

We selected patients who developed fever during their ICU stay after intra-abdominal surgery from the MIMIC-IV and MIMIC-IV-Note databases. After determining the optimal number of clusters, we classified these selected patients into distinct clustered groups using cluster analysis, and analyzed the characteristics of each group. MIMIC-IV and MIMIC-VI-Note are open databases where the Institutional Review Board of the Beth Israel Deaconess Medical Center has waived informed consent and approved sharing of research data. Given that this is a retrospective analysis of an open database, the authors of this study agreed that IRB approval was unnecessary (No. 2025GR0202).

Data Source and Patient Selection

All data for this study was obtained from MIMIC-IV database version 2.2 and MIMIC-IV-Note database version 2.2, which were the most recent versions at the time the study [3]. In these databases, all surgeries are stored as standardized codes according to the International Classification of Diseases (ICD) code systems — the ICD-9 Clinical Modifications (ICD-9-CM) and the ICD-10 Procedure Coding Systems (ICD-10-PCS). We selected patients who underwent intra-abdominal surgery during their hospital stay and stayed in the ICU postoperatively. Procedure codes used to select patients who underwent intra-abdominal surgery are provided in Table 1. However, non-surgical interventions, extra-abdominal surgeries, and transplant surgeries — which might involve immunosuppressants that could affect fever courses — were excluded even if they had the codes listed in Table 1.

In cases where a patient underwent multiple surgeries during admission, only the last surgery was included in the study to minimize the influence of fever caused by earlier surgeries during observation. Instead, if there was a prior surgery before the included procedure during the admission period, we marked the selected surgery as a reoperation, and this was evaluated as a separate variable to assess its impact on postoperative fever.

From the extracted intra-abdominal surgery patients, we further selected patients who developed postoperative fever. Postoperative fever was defined as an elevation of body temperature to 38 °C or higher for 2 consecutive days or 39 °C or higher at least once after surgery [7]. However, we excluded patients who ever had fever in the 3 days before surgery or beyond 14 days (2 weeks) postoperatively, as it was unclear whether the fever was attributable to postoperative status.

All patient data variables were extracted from the MIMIC-IV and MIMIC-IV-Note databases, including demographics, underlying conditions, used medications at fever occurrence, type of surgery (emergency, reoperation) and site of surgery, vital signs, Glasgow coma scale (GCS), laboratory test results, positivity of microbiology cultures, chest x-ray findings, ratio of arterial oxygen partial pressure to fractional inspired oxygen (P/F ratio), ventilator use, Sequential Organ Failure Assessment (SOFA) score and its subscores, and patient outcomes (Table 2). Vital signs, GCS, and laboratory results were obtained and the P/F ratio and SOFA scores were calculated from data on the date of the postoperative fever onset. Microbiology culture results were obtained from data within 3 days before and after the onset of postoperative fever, because there is a considerable time delay between test execution and result reporting and the test is performed intermittently at intervals of several days. For some data which were not stored as codes or numbers in the MIMIC-IV database, such as underlying diseases, the emergency of surgery, and chest x-ray findings, we reviewed discharge summary notes and radiologic notes in the MIMIC-IV-Note database to find this information. When a surgery covered multiple organs, each surgical site was marked on the database, allowing for potential overlap in the marking of surgical sites.

Cluster Analysis

We performed cluster analysis using variables extracted from the databases (Table 2). During cluster analysis, patient outcome variables such as length of stay in the ICU or hospital or death after fever occurrence at 30 days and 7 days were not included as parameters. These variables were included after cluster analysis only to summarize patient characteristics and compare the outcomes of the two groups. We employed GMM [8] to cluster patients. This is a widely used clustering model that is especially suited to handling clusters with diverse distributions, making it ideal for heterogeneous medical data.

The number of clusters was determined using the silhouette method [9] by selecting the number of clusters that yielded the highest silhouette score, indicating the best clustering quality. The silhouette score measures how similar an object is to its own cluster compared to other clusters, with higher values indicating better-defined and well-separated clusters. To effectively visualize the distribution of patient characteristics, principal component analysis (PCA) [10] was used. PCA simplifies numerous complex and multidimensional variables by reducing correlated variables into a smaller number of newly-generated component variables, facilitating the visualization and interpretation of data distribution. We reconstructed patient data variables into five principal component variables and utilized the two most influential variables among them to visualize the distribution of patients in a two-dimensional space using PCA.

Statistical Analysis of Differences between Groups

After clustering the patients into two groups, we assessed differences between variables across groups using classical statistical methods. We used the t-test to assess the significance of differences in continuous variables between the two groups, while we used chi-square analysis to assess the significance of differences in categorical variables between groups. A P-value less than 0.05 was considered statistically significant.

Mortality in Early-onset Postoperative Fever Versus Overall Rates

Early-onset postoperative fever (EPF), which was defined as postoperative fever occurring within postoperative day 2 in this study, is considered to be related to a postoperative reaction rather than an infectious fever [11]. Therefore, EPF is considered to be relatively low risk. To confirm whether this applies to postoperative patients in the ICU, the mortality rate of EPF was calculated separately and checked to see whether it showed a significant difference from the overall mortality rate.

RESULTS

Patient Characteristics

From the MIMIC-IV and MIMIC-IV-Note databases, which contain data from 73,181 ICU stays involving 50,920 patients, a total of 2,858 ICU stays following intra-abdominal surgery were identified. Among these, 331 ICU stays (11.6%) involving 330 patients developed postoperative fever (Figure 1). Overall mean age at surgery was 60.6±16.7 years, with a male predominance (61.0%). Average onset of fever was 3.5±3.9 days after surgery. Common underlying conditions included hypertension (53.5%) and diabetes mellitus (20.8%). Seventy-five patients (22.7%) of patients received vasopressor support at the time of fever. Additionally, the majority of surgeries were emergency procedures (77.0%), with reoperations accounting for 27.8% of cases. The most common surgery site was the small bowel (37.2%) followed by other sites (36.6%), which were mostly exploration or ancillary surgeries such as adhesiolysis.

The proportion of atelectasis and pneumonia in the population was 13.3% and 10.9%, respectively. At the time of fever occurrence, 63.7% of the patients required ventilator support. Mean SOFA score across the cohort was 5.3±3.6. Average hospital length of stay was 27.2±24.6 days, while the average ICU stay was 11.1±11.9 days. The 30-day mortality rate following postoperative fever was 16.9%, and the 7-day mortality rate was 9.7% (Table 3).

Cluster Analysis

Patient distribution was visualized on a two-dimensional plane using PCA (Figure 2). The visualization revealed two distinct patterns: a sparse distribution primarily located on the left side and a denser distribution on the right side, indicating the presence of heterogeneous clustering tendencies among postoperative fever patients. To determine the optimal number of clusters, a silhouette score analysis was conducted, as shown in Figure 3. The analysis evaluated clustering solutions ranging from two to eight clusters. The silhouette score was highest when the number of clusters was set to two, indicating that dividing the patients into two groups provided the most appropriate and distinct classification. Based on this analysis, patients were classified into two distinct groups—group A and group B—based on clinical attributes using GMM cluster analysis (Figure 4A).

Comparison of Cluster Results and the Distribution of Deceased Patients

Figure 4B and C depict the distribution of patients who died on the 7th and 30th day after developing a postoperative fever. These figures demonstrate a substantial overlap between the identified clusters (Figure 4A) and mortalities (Figures 4B and C), highlighting the effectiveness of the clustering method in predicting patient prognosis.

Statistical Analysis of Differences between Groups

The characteristics of the two clustered groups—groups A and B—are summarized in Table 4. Group A included 121 (36.6%) patients who were older, whereas group B included 210 (63.4%) patients who were relatively younger (65.2±14.4 vs. 58.0±17.4 years). There were no significant differences in underlying diseases and use of antipyretics or antibiotics between the two groups. By contrast, there were significant differences in vasopressor use (57.0% vs. 2.9%). There were no significant differences in the type of surgery, such as emergency (78.5% vs. 76.2%) or reoperation (33.9% vs. 24.3%). Regarding surgery site, group A had significantly fewer stomach surgeries (8.3% vs. 17.1%) and small bowel surgeries (46.3% vs. 31.9%) than group B, while no significant differences were observed at other sites. With regard to vital signs, there were no significant differences in maximum body temperature between the two groups, whereas group A had significantly lower systolic, diastolic, and mean arterial blood pressures than group B. In laboratory test results, all variables except white blood cell (WBC) count, which had a marginal p-value of 0.06, showed significant differences between the two groups. There were no significant differences in positivity of microbiologic cultures or chest x-ray findings between the groups. However, P/F ratio (179.4±93.5 vs. 211.7±100.3), ventilator use (76.0% vs. 53.3%), and SOFA score (8.7±2.9 vs. 3.3±2.1) and its subscores except central nervous system (CNS) subscore were significantly different between groups A and B, respectively.

Length of hospital stay and ICU stay showed no differences between the two groups. However, mortalities after postoperative fever showed significant differences both at 30 days (33.1% vs. 7.6%) and 7 days (21.5% vs. 2.9%) after fever. This result was similar when mortalities were calculated for patients who had early-onset fever only (30 days after fever: 23.9% vs. 1.8%, 7 days after fever: 16.4% vs. 0), as shown in Table 4.

Comparison of Mortality in EPF Versus Overall Rates

When comparing EPF patients with the overall population, the mortality rates of EPF patients were always lower regardless of clustering or the time point of mortality assessment (Table 5). However, when it was evaluated using a chi-square test, significance was found at only two time points: the overall 30-day mortality rate of EPF patients (10.2 % vs. 16.9%, P=0.042) and the 30-day mortality rate of EPF patient in group B (1.8% vs. 7.6%, P=0.034).

DISCUSSION

Postoperative fever is now recognized to be associated with various pathophysiological factors rather than just simple factors such as atelectasis or pneumonia [12-15]. Therefore, it is essential to evaluate various aspects of these patients’ conditions and determine what actions to take. This is particularly critical for patients staying in the ICU after surgery, as these patients have a severe clinical status and the potential for rapid disease progression.

Cluster analysis evaluates multiple features of patients and divides them into groups without human bias thereby facilitating rapid assessment of a patient’s condition and enabling timely decision-making regardless of the physician’s experience level. In this study, we classified patients into two distinct groups using GMM cluster analysis. Through this classification, we could divide postoperative fever patients in the ICU into high-risk patients and low-risk patients. Using this method, physicians can confidently perform invasive examinations and procedures on high-risk patients to prevent disease progression, as well as minimize unnecessary invasive interventions in low-risk patients.

We could find various significant differences between the two groups were observed in several categories including surgical site (stomach and small bowel surgeries), vital signs (systolic, diastolic, and mean arterial blood pressures), laboratory test results (all laboratory test results except WBC count), and SOFA scores (excluding CNS scores). Age, P/F ratio, and rates of ventilator use and vasopressor use were also significantly different between the two groups.

Group A had a number of unfavorable conditions compared to group B. First, group A consisted of older patients with a higher SOFA score, suggesting reduced physiological reserve. Secondly, group A had lower pH and higher lactate levels in laboratory results, as well as lower blood pressures along with elevated SOFA scores, suggesting a higher incidence of septic conditions. Thirdly, higher ventilator use, lower P/F ratio, and elevated respiratory SOFA scores indicated that group A had a greater respiratory burden than group B. Collectively, these factors appear to contribute to the dramatic difference in survival rates between group A and group B. Cluster analysis effectively differentiated the two groups by comprehensively analyzing these variables.

Meanwhile, the mortalities of EPF patients were not significantly lower than those of the overall population, except for overall patients and group B patients 30 days after fever (Table 5). This challenges the traditional belief that EPF is a general physiological reaction to surgery and is a low risk condition [1,11]. It suggests, in critically ill patients, that EPF should not be ignored simply because it occurred soon after surgery, particularly when accompanied by hemodynamic instability or elevated SOFA scores.

This study has several limitations. First, this study relied solely on MIMIC-IV databases, which essentially contain retrospective data from a single institute. This may have introduced selection bias and limits the generalizability of the results to other populations or healthcare settings. Second, this study had a relatively small sample size, with a focus on abdominal surgery, preventing generalization of the results to all postoperative fevers. Third, analysis was based on routinely collected clinical data, which may lack certain detailed information such as the specific sources of infection or the exact causes of fever. Fourth, the absence of detailed surgical information, such as surgery record, total surgery time, and estimated bleeding loss, limited the depth of analysis on surgical interventions. Lastly, while cluster analysis offers valuable insights, it does not establish causal relationships between patient characteristics and outcomes.

This study was limited to abdominal surgeries based on data from a single institution; therefore, analysis of data from other institutions and different types of surgery may not result in identification of the same two clusters identified in the present study. However, by applying the clustering method used in this study, physicians can create optimized clustered groups tailored to the specific characteristics of their institution. This approach can aid in the development of customized treatment strategies designed to meet the unique needs of individual patients.

In conclusion, machine learning-based cluster analysis can effectively stratify ICU patients with postoperative fever and allow individualized risk assessment and management. In ICU patients, postoperative fever, even if it is early-onset, should be considered a serious warning sign especially if it is accompanied by hemodynamic instability or high SOFA scores. Incorporating clustering techniques into clinical practice can improve patient outcomes by facilitating the early identification of high-risk patients, while allows timely therapeutic interventions and supports the development of personalized treatment strategies.

KEY MESSAGES

▪ Cluster analysis can effectively classify intensive care unit (ICU) patients with postoperative fever into high-risk and low-risk groups.

▪ Clustering ICU patients with postoperative fever helps determine whether aggressive management or conservative treatment is more appropriate.

▪ In postoperative ICU patients, it is important to be alert for early-onset fever if the patient shows hemodynamic instability or has an elevated Sequential Organ Failure Assessment (SOFA) score.

NOTES

CONFLICT OF INTEREST

No potential conflict of interest relevant to this article was reported.

FUNDING

None.

ACKNOWLEDGMENTS

None.

AUTHOR CONTRIBUTIONS

Conceptualization: SML, HS. Data curation: SML. Formal analysis: SML. Methodology: SML, HS. Project administration: HS. Visualization: SML. Writing – original draft: SML. Writing – review & editing: HJS. All authors read and agreed to the published version of the manuscript.

Figure 1.

Patient selection flowchart illustrating the exclusion process for intensive care unit (ICU) patients with postoperative fever. Of 73,181 ICU stays (50,920 patients), 2,853 ICU stays after intra-abdominal surgery (2,718 patients) were selected. Among these, 331 ICU stays (330 patients) with postoperative fever cases were included in this study.

Figure 2.

Visualization of the total patient data distribution using principal component analysis (PCA). X-axis and y-axis represent the first two principal components, PC1 and PC2, respectively, and variables newly-generated through PCA can capture the most significant patterns in the data. Total patient data was mapped onto this two-dimensional plane before performing cluster analysis. Although the plot does not clearly show distinct group separations, a more dispersed distribution on the left side and a denser distribution on the right side were evident. This pattern suggests the presence of two potential subgroups in the data.

Figure 3.

Evaluation of cluster quality using silhouette scores to determine the optimal number of clusters. The x-axis represents the number of clusters, and the y-axis indicates the silhouette score. This shows evaluation of clustering into 2 to 8 clusters. Higher silhouette scores suggest better-defined clusters, indicating that clustering the patients into two groups is the most appropriate.

Figure 4.

Visualization of patient data distribution. (A) Two-dimensional visualization of clustered patient groups using principal component analysis (PCA) after cluster analysis. Data were divided into two clusters, representing distinct patient subgroups identified through the analysis. (B) Visualization of actual survival distribution 30 days postoperatively after the onset of fever, projected onto a two-dimensional plane using PCA. (C) Visualization of actual survival distribution at 7 days postoperatively after the onset of fever, projected onto a two-dimensional plane using PCA.

Table 1.

Procedure code patterns for intra-abdominal surgeries included in this study

Included surgical site	Included procedure code pattern
Included surgical site	ICD-9-CM	ICD-10-PCS
Stomach	43-, 44-	0D6-, 0D7-
Small bowel	Small bowel surgeries in 45-, 46- (manually selected)	0D8-, 0D9-, 0DA-, 0DB-, 0DC-
Large bowel	Large bowel surgeries in 45-, 46- and 48- (manually selected)	0DE-, 0DF-, 0DG-, 0DH-, 0DK-, 0DL-, 0DM-, 0DN-, 0DP-
Appendix	47-	0DJ-
Liver	50-	0F0-, 0F1-, 0F2-
Gall bladder	510-	0F4-
Duct (bile/pancreas)	Bile or pancreatic surgeries in 51-	0F5-, 0F6-, 0F7-, 0F8-, 0F9-, 0FD-, 0FF-
Pancreas	52-	0FG-
Other	Remaining codes between 43- and 52-	Remaining codes in 0D- and 0F-
Excluded procedures despite matching above code patterns	Endoscopic procedures
	Catheter management (including insertion, removal, reposition) interventions
	Biopsy
	Embolectomy
	Percutaneous surgery
	Abdominal wall surgery
	Hernia surgery
	Anal surgery
	Organ transplantation

ICD: International Classification of Diseases; ICD-9-CM: ICD-9 Clinical Modifications; ICD-10-PCS: ICD-10 Procedure Coding Systems.

Table 2.

Variables used in the study

Variable
Demographics, fever onset, and ventilator use
- Age at surgery, sex, days of fever onset after surgery, early-onset postoperative fever
Underlying conditions
- Hypertension, diabetes mellitus, COPD, chronic heart failure, atrial fibrillation, end-stage renal disease
Medications before fever occurrence
- Antipyretics, antibiotics, vasopressors
Type of surgery
- Emergency, reoperation
Site of surgery
- Stomach, small bowel, large bowel, appendix, liver, gall bladder, bile duct, pancreas, any other sites
SOFA score
- SOFA total score and subscores (respiration, coagulation, liver, cardiovascular, CNS, renal)
Vital signs, GCS
- Body temperature, heart rate, systolic BP, diastolic BP, MAP, O₂ saturation, blood glucose level, GCS
Laboratory test results
- pH, Lactate, WBC count, hemoglobin, platelets, prothrombin time, partial thromboplastin time, total bilirubin, serum creatinine
Positivity in microbiology cultures
- Blood culture, respiratory culture, urine culture, other cultures
Chest x-ray findings
- Atelectasis, pneumonia
P/F ratio
- Ventilator use
Patient outcomes
- Length of hospital stay, length of ICU stay, death within 30 days after fever, death within 7 days after fever, death within 30 days after early-onset fever, death within 7 days after early-onset fever

COPD: chronic obstructive pulmonary disease; SOFA: Sequential Organ Failure Assessment; CNS: central nervous system; GCS: Glasgow coma scale; BP: blood pressure; MAP: mean arterial pressure; WBC: white blood cell; P/F ratio: ratio of arterial oxygen partial pressure to fractional inspired oxygen; ICU: intensive care unit.

Table 3.

Patient characteristics

Variable	Value (n=331)
Demographics
Age at surgery (yr)	60.6±16.7
Male sex	202 (61.0)
Day of fever onset after surgery	3.5±3.0
Early-onset postoperative fever	215 (65.0)
Underlying conditions
Hypertension	177 (53.5)
Diabetes mellitus	69 (20.8)
COPD	30 (9.1)
Chronic heart failure	17 (5.1)
Atrial fibrillation	28 (8.5)
End-stage renal disease	14 (4.2)
Medications at fever occurrence
Antipyretics	223 (67.4)
Antibiotics	259 (78.2)
Vasopressors	75 (22.7)
Type of surgery
Emergency	255 (77.0)
Reoperation	92 (27.8)
Site of surgery
Stomach	46 (13.9)
Small bowel	123 (37.2)
Large bowel	92 (27.8)
Appendix	9 (2.7)
Liver	26 (7.9)
Gall bladder	14 (4.2)
Bile duct	39 (11.8)
Pancreas	25 (7.6)
Any other sites	121 (36.6)
Vital signs, GCS
Body temperature, maximum (℃)	38.6±0.5
Heart rate, mean (beats/min)	101.1±14.5
Systolic BP, maximum (mm Hg)	150.6±23.3
Systolic BP, minimum (mm Hg)	90.2±19.0
Diastolic BP, maximum (mm Hg)	81.3±18.6
Diastolic BP, minimum (mm Hg)	46.5±11.6
MAP, mean (mm Hg)	78.0±11.6
O₂ saturation, minimum (%)	92.0±5.3
Blood glucose level, mean	140.3±47.4
GCS, minimum	13.9±2.4
Laboratory test result
pH, minimum	7.4±0.1
Lactate, maximum (mmol/L)	2.6±2.0
WBC count, maximum (x10³/μl)	14.9±8.2
Hemoglobin, minimum (g/dl)	9.2±1.6
Platelets, minimum (x10³/μl)	221.3±143.3
Prothrombin time, maximum (sec)	16.7±5.5
Partial thromboplastin time (sec)	41.5±22.7
Total bilirubin, maximum (mg/dl)	3.5±5.4
Serum creatinine, maximum (mg/dl)	1.5±1.2
Positivity in microbiology cultures
Blood culture	30 (9.1)
Respiratory culture	88 (26.6)
Urine culture	31 (9.4)
Other cultures	35 (10.6)
Chest x-ray findings
Atelectasis	44 (13.3)
Pneumonia	36 (10.9)
P/F ratio	196.8±98.3
Ventilator use	204 (61.6)
SOFA score
Total score	5.3±3.6
Respiration	1.8±1.5
Coagulation	0.6±0.9
Liver	1.3±1.3
Cardiovascular	1.6±1.4
CNS	0.5±1.0
Renal	1.0±1.3
Patient outcome
Length of hospital stay (day)	27.2±24.6
Length of ICU stay (day)	11.1±11.9
30-Day mortality after fever (%)	56 (16.9)
7-Day mortality after fever (%)	32 (9.7)
30-Day mortality after early-onset fever (%)	18 (10.2)
7-Day mortality after early-onset fever (%)	11 (6.3)

Values are presented as mean±standard deviation or number (%). Early-onset postoperative fever is defined as postoperative fever that occurs within postoperative day 2. Medications, vital signs, GCS, laboratory test results, ventilator use, SOFA score, were checked at fever occurrence day.

COPD: chronic obstructive pulmonary disease; GCS: Glasgow coma scale; BP: blood pressure; MAP: mean arterial pressure; WBC: white blood cell; P/F ratio: ratio of arterial oxygen partial pressure to fractional inspired oxygen; SOFA: Sequential Organ Failure Assessment; CNS: central nervous system; ICU: intensive care unit.

Table 4.

Total sums (percentage), average values, and P-values of variables for each group

Variable	Group A (n= 121)	Group B (n = 210)	P-value
Demographics
Age at surgery (yr)	65.2±14.4	58.0±17.4	<0.001
Male sex	73 (60.3)	129 (61.4)	0.936
Days of fever onset after surgery	3.4±3.1	3.5±3.0	0.788
Early-onset postoperative fever	67 (55.4)	109 (51.9)	0.621
Underlying conditions
Hypertension	67 (55.4)	110 (52.4)	0.442
Diabetes mellitus	29 (24.0)	40 (19.0)	0.274
COPD	11 (9.1)	19 (9.0)	1.000
Chronic heart failure	8 (6.6)	9 (4.3)	0.457
Atrial fibrillation	11 (9.1)	17 (8.1)	0.836
End-stage renal disease	7 (5.8)	7 (3.3)	0.392
Medications before fever occurrence
Antipyretics	76 (62.8)	147 (70.0)	0.222
Antibiotics	96 (79.3)	163 (77.6)	0.820
Vasopressors	69 (57.0)	6 (2.9)	<0.001
Type of surgery
Emergency	95 (78.5)	160 (76.2)	0.324
Reoperation	41 (33.9)	51 (24.3)	0.080
Site of surgery
Stomach	10 (8.3)	36 (17.1)	0.037
Small bowel	56 (46.3)	67 (31.9)	0.013
Large bowel	32 (26.4)	60 (28.6)	0.773
Appendix	4 (3.3)	5 (2.4)	0.883
Liver	10 (8.3)	16 (7.6)	1.000
Gall bladder	6 (5.0)	8 (3.8)	0.828
Bile duct	15 (12.4)	24 (11.4)	0.931
Pancreas	11 (9.1)	14 (6.7)	0.557
Any other sites	42 (34.7)	79 (37.6)	0.681
Vital signs, GCS
Body temperature, maximum (℃)	38.7±0.5	38.6±0.5	0.148
Heart rate, mean (beats/min)	100.5±14.5	101.5±14.6	0.604
Systolic BP, maximum (mm Hg)	138.7±18.1	158.2±23.1	<0.001
Systolic BP, minimum (mm Hg)	77.1±12.5	98.4±17.8	<0.001
Diastolic BP, maximum (mm Hg)	70.6±14.0	88.1±18.0	<0.001
Diastolic BP, minimum (mm Hg)	39.8±7.1	50.6±12.0	<0.001
MAP, mean (mm Hg)	69.7±6.1	83.1±11.3	<0.001
O₂ saturation, minimum (%)	91.3±6.6	92.5±4.2	0.071
Blood glucose level, mean	148.0±63.6	135.4±32.6	0.058
GCS, minimum	13.8±2.7	14.0±2.2	0.502
Laboratory test results
pH, minimum	7.3±0.1	7.4±0.1	<0.001
Lactate, maximum (mmol/L)	3.4±2.3	1.9±1.3	<0.001
WBC count, maximum (x10³/μl)	16.0±8.6	14.2±8.0	0.064
Hemoglobin, minimum (g/dl)	8.9±1.7	9.4±1.6	0.015
Platelets, minimum (x10³/μl)	182.0±139.2	243.5±141.2	<0.001
Prothrombin time, maximum (sec)	18.7±7.0	15.3±3.6	<0.001
Partial thromboplastin time (sec)	45.4±23.3	38.6±22.0	0.024
Total bilirubin, maximum (mg/dl)	5.2±6.9	2.1±3.2	<0.001
Serum creatinine, maximum (mg/dl)	2.0±1.4	1.2±0.9	<0.001
Positivity in microbiology cultures
Blood culture	10 (8.3)	20 (9.5)	0.853
Respiratory culture	38 (31.4)	50 (23.8)	0.168
Urine culture	11 (9.1)	20 (9.5)	1.000
Other cultures	15 (12.4)	20 (9.5)	0.527
Chest x-ray findings
Atelectasis	20 (16.5)	24 (11.4)	0.251
Pneumonia	16 (13.2)	20 (9.5)	0.391
P/F ratio	179.4±93.5	211.7±100.3	0.011
Ventilator use	92 (76.0)	112 (53.3)	<0.001
SOFA score
Total score	8.7±2.9	3.3±2.1	<0.001
Respiration	2.1±1.5	1.5±1.4	<0.001
Coagulation	1.1±1.2	0.3±0.6	<0.001
Liver	1.7±1.4	0.9±1.1	<0.001
Cardiovascular	2.8±1.4	0.8±0.7	<0.001
CNS	0.5±1.0	0.5±0.9	0.955
Renal	1.6±1.5	0.6±1.0	<0.001
Patient outcomes
Length of hospital stay (day)	29.0±23.6	26.2±25.1	0.316
Length of ICU stay (day)	12.8±11.8	10.2±11.9	0.057
30-Day mortality after fever (%)	40 (33.1)	16 (7.6)	<0.001
7-Day mortality after fever (%)	26 (21.5)	6 (2.9)	<0.001
30-Day mortality after early-onset fever (%)	16 (23.9)	2 (1.8)	<0.001
7-Day mortality after early-onset fever (%)	11 (16.4)	0	<0.001

Table 5.

Comparison between early-onset mortalities and overall mortalities

Mortality	Time points	Early-onset postoperative fever only	Overall population	P-value
Overall mortality (%)	30 Days	10.2	16.9	0.042
Overall mortality (%)	7 Days	6.3	9.7	0.189
Group A mortality (%)	30 Days	23.9	33.1	0.188
Group A mortality (%)	7 Days	16.4	21.5	0.402
Group B mortality (%)	30 Days	1.8	7.6	0.034
Group B mortality (%)	7 Days	0	2.9	0.075

Early-onset postoperative fever is defined as postoperative fever that occurs within postoperative day 2.

REFERENCES

1. Narayan M, Medinilla SP. Fever in the postoperative patient. Emerg Med Clin North Am 2013;31:1045-58.Article PubMed
2. Stephenson C, Mohabbat A, Raslau D, Gilman E, Wight E, Kashiwagi D. Management of common postoperative complications. Mayo Clin Proc 2020;95:2540-54.Article PubMed
3. Johnson A, Bulgarelli L, Pollard T, Horng S, Celi LA, Mark R. MIMIC-IV (version 2.2). PhysioNet; 2023.
4. Johnson AE, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data 2023;10:1.Article PubMed PMC PDF
5. Goldberger AL, Amaral LA, Glass L, Hausdorff JM, Ivanov PC, Mark RG, et al. PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. Circulation 2000;101:E215-20.Article PubMed
6. Johnson A, Pollard T, Horng S, Celi LA, Mark R. MIMIC-IV-Note: deidentified free-text clinical notes (v2.2). PhysioNet; 2023.
7. Abdelmaseeh TA, Azmat CE, Oliver TI. Postoperative fever [Internet]. StatPearls Publishing; 2024 [cited 2025 Jan 10]. Available from: http://www.ncbi.nlm.nih.gov/books/NBK482299/
8. Bishop CM. Pattern recognition and machine learning. Springer; 2006.
9. Rousseeuw PJ. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. J Comput Appl Math 1987;20:53-65.Article
10. Halko N, Martinsson PG, Tropp JA. Finding structure with randomness: probabilistic algorithms for constructing approximate matrix decompositions. SIAM Rev 2011;53:217-88.Article
11. Lim L, Lee J, Hwang SY, Lee H, Oh SY, Kang C, et al. Early postoperative fever and atelectasis in patients undergoing upper abdominal surgery. J Am Coll Surg 2023;237:606-13.Article PubMed
12. Nohra E, Appelbaum RD, Farrell MS, Carver T, Jung HS, Kirsch JM, et al. Fever and infections in surgical intensive care: an American Association for the Surgery of Trauma Critical Care Committee clinical consensus document. Trauma Surg Acute Care Open 2024;9:e001303. Article PubMed PMC
13. Caballero-Herrero MJ, Jumilla E, Buitrago-Ruiz M, Valero-Navarro G, Cuevas S. Role of Damage-Associated Molecular Patterns (DAMPS) in the postoperative period after colorectal surgery. Int J Mol Sci 2023;24:3862.Article PubMed PMC
14. Crompton JG, Crompton PD, Matzinger P. Does atelectasis cause fever after surgery?: putting a damper on dogma. JAMA Surg 2019;154:375-6.Article PubMed
15. Bassetti M, Eckmann C, Giacobbe DR, Sartelli M, Montravers P. Post-operative abdominal infections: epidemiology, operational definitions, and outcomes. Intensive Care Med 2020;46:163-72.Article PubMed PDF

Figure & Data

References

Citations

Citations to this article as recorded by

PubReader
ePub Link

Cite

CITE: export Copy Format; Close

Download Citation

Download a citation file in RIS format that can be imported by all major citation management software, including EndNote, ProCite, RefWorks, and Reference Manager.

Format:

RIS — For EndNote, ProCite, RefWorks, and most other reference management software
BibTeX — For JabRef, BibDesk, and other BibTeX-specific software

Include:

Citation for the content below
Citation and abstract for the content below

Classification of postoperative fever patients in the intensive care unit following intra-abdominal surgery: a machine learning-based cluster analysis using the Medical Information Mart for Intensive Care (MIMIC)-IV database, developed in the United States

DOI: https://doi.org/10.4266/acc.004464

XML Download

Figure

Related articles

Early bronchoscopy in severe pneumonia patients in intensive care unit: insights from the Medical Information Mart for Intensive Care-IV database analysis

Figure 1. Patient selection flowchart illustrating the exclusion process for intensive care unit (ICU) patients with postoperative fever. Of 73,181 ICU stays (50,920 patients), 2,853 ICU stays after intra-abdominal surgery (2,718 patients) were selected. Among these, 331 ICU stays (330 patients) with postoperative fever cases were included in this study.

Figure 2. Visualization of the total patient data distribution using principal component analysis (PCA). X-axis and y-axis represent the first two principal components, PC1 and PC2, respectively, and variables newly-generated through PCA can capture the most significant patterns in the data. Total patient data was mapped onto this two-dimensional plane before performing cluster analysis. Although the plot does not clearly show distinct group separations, a more dispersed distribution on the left side and a denser distribution on the right side were evident. This pattern suggests the presence of two potential subgroups in the data.

Figure 3. Evaluation of cluster quality using silhouette scores to determine the optimal number of clusters. The x-axis represents the number of clusters, and the y-axis indicates the silhouette score. This shows evaluation of clustering into 2 to 8 clusters. Higher silhouette scores suggest better-defined clusters, indicating that clustering the patients into two groups is the most appropriate.

Figure 4. Visualization of patient data distribution. (A) Two-dimensional visualization of clustered patient groups using principal component analysis (PCA) after cluster analysis. Data were divided into two clusters, representing distinct patient subgroups identified through the analysis. (B) Visualization of actual survival distribution 30 days postoperatively after the onset of fever, projected onto a two-dimensional plane using PCA. (C) Visualization of actual survival distribution at 7 days postoperatively after the onset of fever, projected onto a two-dimensional plane using PCA.

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Included surgical site	Included procedure code pattern
Included surgical site	ICD-9-CM	ICD-10-PCS
Stomach	43-, 44-	0D6-, 0D7-
Small bowel	Small bowel surgeries in 45-, 46- (manually selected)	0D8-, 0D9-, 0DA-, 0DB-, 0DC-
Large bowel	Large bowel surgeries in 45-, 46- and 48- (manually selected)	0DE-, 0DF-, 0DG-, 0DH-, 0DK-, 0DL-, 0DM-, 0DN-, 0DP-
Appendix	47-	0DJ-
Liver	50-	0F0-, 0F1-, 0F2-
Gall bladder	510-	0F4-
Duct (bile/pancreas)	Bile or pancreatic surgeries in 51-	0F5-, 0F6-, 0F7-, 0F8-, 0F9-, 0FD-, 0FF-
Pancreas	52-	0FG-
Other	Remaining codes between 43- and 52-	Remaining codes in 0D- and 0F-
Excluded procedures despite matching above code patterns	Endoscopic procedures
	Catheter management (including insertion, removal, reposition) interventions
	Biopsy
	Embolectomy
	Percutaneous surgery
	Abdominal wall surgery
	Hernia surgery
	Anal surgery
	Organ transplantation

Variable
Demographics, fever onset, and ventilator use
- Age at surgery, sex, days of fever onset after surgery, early-onset postoperative fever
Underlying conditions
- Hypertension, diabetes mellitus, COPD, chronic heart failure, atrial fibrillation, end-stage renal disease
Medications before fever occurrence
- Antipyretics, antibiotics, vasopressors
Type of surgery
- Emergency, reoperation
Site of surgery
- Stomach, small bowel, large bowel, appendix, liver, gall bladder, bile duct, pancreas, any other sites
SOFA score
- SOFA total score and subscores (respiration, coagulation, liver, cardiovascular, CNS, renal)
Vital signs, GCS
- Body temperature, heart rate, systolic BP, diastolic BP, MAP, O₂ saturation, blood glucose level, GCS
Laboratory test results
- pH, Lactate, WBC count, hemoglobin, platelets, prothrombin time, partial thromboplastin time, total bilirubin, serum creatinine
Positivity in microbiology cultures
- Blood culture, respiratory culture, urine culture, other cultures
Chest x-ray findings
- Atelectasis, pneumonia
P/F ratio
- Ventilator use
Patient outcomes
- Length of hospital stay, length of ICU stay, death within 30 days after fever, death within 7 days after fever, death within 30 days after early-onset fever, death within 7 days after early-onset fever

Variable	Value (n=331)
Demographics
Age at surgery (yr)	60.6±16.7
Male sex	202 (61.0)
Day of fever onset after surgery	3.5±3.0
Early-onset postoperative fever	215 (65.0)
Underlying conditions
Hypertension	177 (53.5)
Diabetes mellitus	69 (20.8)
COPD	30 (9.1)
Chronic heart failure	17 (5.1)
Atrial fibrillation	28 (8.5)
End-stage renal disease	14 (4.2)
Medications at fever occurrence
Antipyretics	223 (67.4)
Antibiotics	259 (78.2)
Vasopressors	75 (22.7)
Type of surgery
Emergency	255 (77.0)
Reoperation	92 (27.8)
Site of surgery
Stomach	46 (13.9)
Small bowel	123 (37.2)
Large bowel	92 (27.8)
Appendix	9 (2.7)
Liver	26 (7.9)
Gall bladder	14 (4.2)
Bile duct	39 (11.8)
Pancreas	25 (7.6)
Any other sites	121 (36.6)
Vital signs, GCS
Body temperature, maximum (℃)	38.6±0.5
Heart rate, mean (beats/min)	101.1±14.5
Systolic BP, maximum (mm Hg)	150.6±23.3
Systolic BP, minimum (mm Hg)	90.2±19.0
Diastolic BP, maximum (mm Hg)	81.3±18.6
Diastolic BP, minimum (mm Hg)	46.5±11.6
MAP, mean (mm Hg)	78.0±11.6
O₂ saturation, minimum (%)	92.0±5.3
Blood glucose level, mean	140.3±47.4
GCS, minimum	13.9±2.4
Laboratory test result
pH, minimum	7.4±0.1
Lactate, maximum (mmol/L)	2.6±2.0
WBC count, maximum (x10³/μl)	14.9±8.2
Hemoglobin, minimum (g/dl)	9.2±1.6
Platelets, minimum (x10³/μl)	221.3±143.3
Prothrombin time, maximum (sec)	16.7±5.5
Partial thromboplastin time (sec)	41.5±22.7
Total bilirubin, maximum (mg/dl)	3.5±5.4
Serum creatinine, maximum (mg/dl)	1.5±1.2
Positivity in microbiology cultures
Blood culture	30 (9.1)
Respiratory culture	88 (26.6)
Urine culture	31 (9.4)
Other cultures	35 (10.6)
Chest x-ray findings
Atelectasis	44 (13.3)
Pneumonia	36 (10.9)
P/F ratio	196.8±98.3
Ventilator use	204 (61.6)
SOFA score
Total score	5.3±3.6
Respiration	1.8±1.5
Coagulation	0.6±0.9
Liver	1.3±1.3
Cardiovascular	1.6±1.4
CNS	0.5±1.0
Renal	1.0±1.3
Patient outcome
Length of hospital stay (day)	27.2±24.6
Length of ICU stay (day)	11.1±11.9
30-Day mortality after fever (%)	56 (16.9)
7-Day mortality after fever (%)	32 (9.7)
30-Day mortality after early-onset fever (%)	18 (10.2)
7-Day mortality after early-onset fever (%)	11 (6.3)

Variable	Group A (n= 121)	Group B (n = 210)	P-value
Demographics
Age at surgery (yr)	65.2±14.4	58.0±17.4	<0.001
Male sex	73 (60.3)	129 (61.4)	0.936
Days of fever onset after surgery	3.4±3.1	3.5±3.0	0.788
Early-onset postoperative fever	67 (55.4)	109 (51.9)	0.621
Underlying conditions
Hypertension	67 (55.4)	110 (52.4)	0.442
Diabetes mellitus	29 (24.0)	40 (19.0)	0.274
COPD	11 (9.1)	19 (9.0)	1.000
Chronic heart failure	8 (6.6)	9 (4.3)	0.457
Atrial fibrillation	11 (9.1)	17 (8.1)	0.836
End-stage renal disease	7 (5.8)	7 (3.3)	0.392
Medications before fever occurrence
Antipyretics	76 (62.8)	147 (70.0)	0.222
Antibiotics	96 (79.3)	163 (77.6)	0.820
Vasopressors	69 (57.0)	6 (2.9)	<0.001
Type of surgery
Emergency	95 (78.5)	160 (76.2)	0.324
Reoperation	41 (33.9)	51 (24.3)	0.080
Site of surgery
Stomach	10 (8.3)	36 (17.1)	0.037
Small bowel	56 (46.3)	67 (31.9)	0.013
Large bowel	32 (26.4)	60 (28.6)	0.773
Appendix	4 (3.3)	5 (2.4)	0.883
Liver	10 (8.3)	16 (7.6)	1.000
Gall bladder	6 (5.0)	8 (3.8)	0.828
Bile duct	15 (12.4)	24 (11.4)	0.931
Pancreas	11 (9.1)	14 (6.7)	0.557
Any other sites	42 (34.7)	79 (37.6)	0.681
Vital signs, GCS
Body temperature, maximum (℃)	38.7±0.5	38.6±0.5	0.148
Heart rate, mean (beats/min)	100.5±14.5	101.5±14.6	0.604
Systolic BP, maximum (mm Hg)	138.7±18.1	158.2±23.1	<0.001
Systolic BP, minimum (mm Hg)	77.1±12.5	98.4±17.8	<0.001
Diastolic BP, maximum (mm Hg)	70.6±14.0	88.1±18.0	<0.001
Diastolic BP, minimum (mm Hg)	39.8±7.1	50.6±12.0	<0.001
MAP, mean (mm Hg)	69.7±6.1	83.1±11.3	<0.001
O₂ saturation, minimum (%)	91.3±6.6	92.5±4.2	0.071
Blood glucose level, mean	148.0±63.6	135.4±32.6	0.058
GCS, minimum	13.8±2.7	14.0±2.2	0.502
Laboratory test results
pH, minimum	7.3±0.1	7.4±0.1	<0.001
Lactate, maximum (mmol/L)	3.4±2.3	1.9±1.3	<0.001
WBC count, maximum (x10³/μl)	16.0±8.6	14.2±8.0	0.064
Hemoglobin, minimum (g/dl)	8.9±1.7	9.4±1.6	0.015
Platelets, minimum (x10³/μl)	182.0±139.2	243.5±141.2	<0.001
Prothrombin time, maximum (sec)	18.7±7.0	15.3±3.6	<0.001
Partial thromboplastin time (sec)	45.4±23.3	38.6±22.0	0.024
Total bilirubin, maximum (mg/dl)	5.2±6.9	2.1±3.2	<0.001
Serum creatinine, maximum (mg/dl)	2.0±1.4	1.2±0.9	<0.001
Positivity in microbiology cultures
Blood culture	10 (8.3)	20 (9.5)	0.853
Respiratory culture	38 (31.4)	50 (23.8)	0.168
Urine culture	11 (9.1)	20 (9.5)	1.000
Other cultures	15 (12.4)	20 (9.5)	0.527
Chest x-ray findings
Atelectasis	20 (16.5)	24 (11.4)	0.251
Pneumonia	16 (13.2)	20 (9.5)	0.391
P/F ratio	179.4±93.5	211.7±100.3	0.011
Ventilator use	92 (76.0)	112 (53.3)	<0.001
SOFA score
Total score	8.7±2.9	3.3±2.1	<0.001
Respiration	2.1±1.5	1.5±1.4	<0.001
Coagulation	1.1±1.2	0.3±0.6	<0.001
Liver	1.7±1.4	0.9±1.1	<0.001
Cardiovascular	2.8±1.4	0.8±0.7	<0.001
CNS	0.5±1.0	0.5±0.9	0.955
Renal	1.6±1.5	0.6±1.0	<0.001
Patient outcomes
Length of hospital stay (day)	29.0±23.6	26.2±25.1	0.316
Length of ICU stay (day)	12.8±11.8	10.2±11.9	0.057
30-Day mortality after fever (%)	40 (33.1)	16 (7.6)	<0.001
7-Day mortality after fever (%)	26 (21.5)	6 (2.9)	<0.001
30-Day mortality after early-onset fever (%)	16 (23.9)	2 (1.8)	<0.001
7-Day mortality after early-onset fever (%)	11 (16.4)	0	<0.001

Mortality	Time points	Early-onset postoperative fever only	Overall population	P-value
Overall mortality (%)	30 Days	10.2	16.9	0.042
Overall mortality (%)	7 Days	6.3	9.7	0.189
Group A mortality (%)	30 Days	23.9	33.1	0.188
Group A mortality (%)	7 Days	16.4	21.5	0.402
Group B mortality (%)	30 Days	1.8	7.6	0.034
Group B mortality (%)	7 Days	0	2.9	0.075

Table 1. Procedure code patterns for intra-abdominal surgeries included in this study

ICD: International Classification of Diseases; ICD-9-CM: ICD-9 Clinical Modifications; ICD-10-PCS: ICD-10 Procedure Coding Systems.

Table 2. Variables used in the study

Table 3. Patient characteristics

Table 4. Total sums (percentage), average values, and P-values of variables for each group

Table 5. Comparison between early-onset mortalities and overall mortalities

Early-onset postoperative fever is defined as postoperative fever that occurs within postoperative day 2.