Prediction Model and Novel Classification System for Complications after Minimally Invasive Hysterectomy.

Data from four national Swedish registries was merged: The Swedish National Quality Register for Gynecological Surgery (GynOp), Swedish National Drug Registry (NDR), Statistics Sweden, and the Swedish National Patient Registry (NPR). Patients who have had a hysterectomy, 1998–2022, with or without concomitant adnexal surgery were identified from the Swedish GynOp Register. GynOp contains preoperative, intraoperative, and postoperative information regarding gynecological surgery since 1997. Data are collected from both the surgeon and the patient. The patient fills in a preoperative questionnaire in order to collect baseline demographic data and also fills in a postoperative questionnaire 8 weeks and 12 months after surgery regarding patient-reported complications. The surgeon confirms or rejects the patient-reported complications in the registry. Demographic variables are filled in at the time of surgery (age, BMI, tobacco use, previous abdominal surgery, previous cesarean section), type of surgery (hysterectomy and/or adnexal surgery), primary incision (abdominal, laparoscopic, vaginal, and conversion to laparoscopy/laparotomy), and indication for surgery. Postoperatively, the surgeon registers surgical data such as uterus weight, blood loss, and operation duration. Although uterus weight was measured postoperatively, uterus size is known preoperatively through clinical examination and ultrasound. The Swedish National Patient Registry contains data on patients admitted to all Swedish hospitals since 1987: date of admission, date of discharge, and diagnoses and procedural codes registered for each admission. Codes are classified according to the International Classification of Diseases, 10th Revision codes. ICD-10 codes for complications as specified below were collected from the patient registry. Indications for hysterectomy were defined as endometriosis N80, prolapse N81.0–81.9, dysmenorrhea N94, cervical dysplasia N87, endometrial hyperplasia N85.0–85.1, menometrorrhagia N92, postmenopausal bleeding N95, and myoma D25. The Prescribed Drug Register was established in 2005 and is maintained by the Swedish National Board of Health and Welfare. Because the register was introduced after the start of the study period, medication data were only available for hysterectomies performed from 2005 onward. It contains data on all prescribed drugs dispensed at pharmacies in Sweden. Pharmaceutical consumption is classified according to the Anatomic Therapeutic Chemical Classification (ATC). Prescriptions of drugs were extracted to identify preexisting comorbidities such as cardiovascular disease or diabetes (3 months prior to surgery) and also to assess postoperative adverse outcome (21 days after surgery). Data regarding drug consumption 3 months prior to surgery were extracted from PDR: A10A insulin, A10B oral antidiabetic agents, B01 antithrombotic agents, C01 cardiac therapy, C02 antihypertensives C03 diuretics, C04 peripheral vasodilators, C05 vasoprotectives, C07 beta-blocking agents, C08 calcium channel blockers, C09 agents acting on the renin-angiotensin system, C10 lipid modifying agents, N05 psycholeptics, N06A antidepressants, N06B psychostimulants, H02AB glucocorticosteroids, L01XC monoclonal antibodies, L03 immunostimulating drugs, L04 immunosuppressing drugs. The following codes were used to identify adverse outcomes 21 days after surgery: J01 antibacterials for systemic use and opioids N02A. Statistics Sweden is a government agency that produces official statistics. From Statistics Sweden, data regarding educational level and country of birth were obtained. Educational level was categorized as comprehensive school, secondary upper level school, and university. The complication composite included any of the following: perioperative complication, perioperative blood loss more than 300 mL, conversion to open surgery, postoperative complication, reoperation, and prescription of antibiotics within 6 weeks postoperatively. All the other patients were defined to have a successful hysterectomy without any deviation from intended course. Intraoperative complications were defined as urinary bladder injury (ICD code: S37.2) (procedural codes: suture of urinary bladder: KCH00, KCH96), ureter injury (ICD code: S37.1) (procedural codes: suture of ureter: KBH00, KBH96), bowel injury (ICD codes: S36.4, S36.5, S36.6) (suture of rectum: JGA60), accidental puncture and laceration during a procedure, not elsewhere classified (T81.2), or vascular complications following a procedure, not elsewhere classified (T81.7). ICD and procedural codes were identified from the Patient Registry. A tick box description of perioperative complications found in GynOp Registry was used. Postoperative complications were defined as tick box code for postoperative complication in Gyn Registry, or presence of Clavien Dindo classification in GynOp. Clavien-Dindo classification is available in the registry from 2016 onward and applies only to postoperative events; perioperative complications were therefore identified using ICD and procedural codes. Postoperative complications of procedures, not elsewhere classified (T81.0), including hematoma, shock, disruption of operation wound, infection, were found in Patient Registry. Other adverse outcomes were defined as presence of prescription of antibiotics 1–21 days postop (ATC code J01 found in the Drug Registry), presence of prescription of opioids 1–21 days postop (ATC code N02A found in Drug Registry), reoperation defined in GynOp Registry and/or as procedural codes LW in Patient Registry. Conversion was identified in the GynOp Registry as a tick box code. The dataset was divided into one development and one validation dataset. This approach ensured an unbiased and balanced split and is appropriate given the large sample size. The model was developed and reported in accordance with the TRIPOD guidelines. From the development dataset, variables associated with complications were identified using univariable logistic regression. For each factor, the optimal format to enter the model was decided (e.g., as class or linear, continuous variables). Possible interactions were tested for. All factors identified to be associated with hysterectomy outcome ( p < 0.2) were initially entered to a first multivariable model. The final model included variables with p value <0.05 only. The Hosmer-Lemeshow test [ 12 ] was used to compare the goodness-of-fit between-different models. The obtained ORs were applied on the validation dataset, and the predicted risk for any complication was estimated for each hysterectomy. For each risk strata, the observed and the estimated risk was compared. The method proposed by DeLong et al. [ 13 ] was used to compute the receiver operating characteristic area under curve with 95% CI. Statistical analyses were performed using IBM SPSS Statistics, version 27.0 (Armonk, NY: IBM Corp, USA), and Gauss (Aptech Systems Inc., AZ, USA).

Data regarding 97,948 hysterectomies performed 1998–2022 were extracted from the Swedish GynOp Registry. After exclusion for missing data regarding uterus weight and BMI, 60,424 hysterectomies were included for analysis. A decline in abdominal and vaginal hysterectomies to the benefit of laparoscopic and robotic hysterectomies was seen during the time period. Open hysterectomies without specification about incision type were classified as Pfannenstiel/Cohen incisions. Vaginal natural orifice transluminal endoscopic surgery ( n = 139) hysterectomies were classified as VH. Table 1 shows patient demographics, intra- and postoperative outcomes depending on mode of hysterectomy, in both open surgery and MIS. Demographic and outcome data varied between patients undergoing open hysterectomy versus MIS and also within the group undergoing MIS. As seen in Table 1 , the characteristics varied significantly by operation type ( p for homogeneity was <0.001 for most demographic groups, patient groups, and outcomes). Type of surgery by group characteristics and peri- and postoperative complications As shown in Table 1 , 53% undergoing LH had any complication qualifying for the composite complication outcome. The corresponding rate for RH was 37.2% and 46.5% for VH. Among patients without any composite complication, the rate of a postoperative prescription of opioids were for LH 14.1%, RH 15.8%, and VH 9.0%. The corresponding rates of a postoperative prescription of opioids among patients with any complication were for LH 17.71%, RH 26.7%, and 14%. Table 2 shows the rate of the composite complications among the different MIS techniques, depending on patient characteristics. The risk of complications were less common in all MIS techniques if the indication for hysterectomy was prolapse and more common if the indication was endometriosis. Complications were more common if the patient had higher BMI, had undergone a previous CS, and decreased with age. Intra-/postoperative complications by type of minimal-invasive operation method Table 3 shows the odds ratio (OR) for any complication among MIS techniques, estimates from crude models and the final multiple model. A strong interaction was found ( p < 0.001) between uterus weight and surgical technique used for hysterectomy on the influence on complication risk. To create a model that took the interaction into account, a set of class variables was designed, combining operation type and uterus weight. The reference class was set to RH operation on uterus weighing ≤300 g (the class with the lowest complication risk). The largest risk of complications was among patients with uterus weight over 1 kg that underwent a VH, OR in the multiple model = 25.5 (95% CI 3.3–199.8), as compared to RH with uterus weight ≤ 300 g. The corresponding OR for women who underwent RH with uterus weight >1 kg was OR = 5.0 (95% CI 2.7–9.4). No other interactions between factors on their impact on complication risk were identified. The risk of complications increased with increasing BMI (OR = 1.90; 95% CI 1.38–2.61) for patients with BMI over 40 as compared to BMI 18.5–25 and decreased with age. The odds for any complications were reduced with increasing age; patients in their 60 s had almost half the odds of complications compared to patients under 50 years old. As both BMI and age were found to be linearly associated (positively and negatively, respectively) with complication risk, these variables were entered as continuous, linear variables in the final model, which improved the goodness of fit. The Hosmer-Lemeshow p value for homogeneity increased from 0.047 to 0.059, for the full model with two linear components as compared to the model with class variables only. The use of drugs for diabetes, cardiovascular disease, or psychiatric disease had no impact on the risk of complications after hysterectomy. OR for any complication a derived from development dataset Estimates from crude models and multiple model, respectively. The multiple model includes all factors listed in the column. a Re-peri-post-op complication, blood loss >300 mL, postoperative antibiotics, or conversion. b Includes all listed factors (factors with p value < 0.2 in the univariate models, excluding factors with p 0.20 in the first multiple model). c Divided into 12 operation type and uterus size classes to account for the significant interaction between operation type and uterus size. The risk of complications increased with increasing rate of previously undergone CS. After one previous CS, compared to parous women without CS, the OR for any complication in the multiple model was = 1.2 (95% CI 1.08–1.34) and after two or more CS, the risk was OR 1.47 (1.26-1–72-9) (data not shown in table). Nulliparity had a protective effect against complications, OR in the multiple model = 0.77 (95% CI 0.66–0.89). The obtained estimates from the final model were applied to the validation dataset. The model was developed in a randomly selected development dataset and validated in an independent dataset, reflecting a predictive modeling approach. Figure 1 shows the comparison between the predicted and the observed rate of complications within each 5% estimated risk strata. The calibration plot ( Fig. 1 ) demonstrates close agreement between predicted and observed complication risks across strata, supporting the model’s calibration. The agreement is strong, with exception of patients with an estimated risk for complications exceeding 90% (but the observed rate in that risk stratum was based on eight persons only). Predictive accuracy of the prediction model for hysterectomy complications. The receiver operating characteristic area under curve yielded a value of 0.65 (95% CI: 0.64–0.66) ( Fig. 2 ). The curve shows the specificity and sensitivity of complications for each individual. The agreement between observed and estimated mean risks was strong, but the overall ability of the model to predict any complication on an individual level was modest. HCCs 1–5 were thereafter defined depending on the rate of composite complications after hysterectomy as HCC 1: 0–20%, HCC 2: 21–40%, HCC 3: 31–60%, HCC 4; 61–80%, and HCC 5: 81–100%. Receiver operating characteristic curve for the prediction model.

Hysterectomy is the most commonly performed major operation in women. Over 400,000 hysterectomies are performed annually in the USA and it is estimated that 30% of women will have had a hysterectomy by age 60 years [ 1 ]. There are multiple techniques on how to perform a hysterectomy, and both patient characteristics and surgeon expertise will influence the likelihood that the hysterectomy will be performed without complications. The approaches to hysterectomy may be broadly categorized into abdominal hysterectomy and minimally invasive surgery (MIS) including VH, vaginal natural orifice transluminal endoscopic surgery, laparoscopic hysterectomy (LH), and robotic hysterectomy (RH) [ 2 ]. The patient characteristics influencing the difficulty level to perform the hysterectomy, what technique to use, level of surgical expertise needed, and the risk of complications include body mass index (BMI) [ 3 ], previous cesarean section [ 4 , 5 ], parity [ 6 ], previous abdominal surgery, indication of surgery, uterus size, smoking, presence of known endometriosis or adhesions, and presence of other comorbidities [ 7 ]. Other potential risk factors for adverse outcome after hysterectomy, albeit less studied, are socioeconomic factors such as educational level and country of birth [ 8 – 11 ]. The variability in patient characteristics, uterus characteristics, and surgical indications makes it difficult to assess and compare study results when new hysterectomy techniques are being introduced or when hysterectomy techniques are being compared. In both clinical practice and in research, a tool to select hysterectomy technique, determining what surgeon expertise is needed, and evaluating and comparing results of hysterectomy studies is missing. The aim of this study is to develop a prediction model for MIS hysterectomy complications, and subsequently defining Hysterectomy Complication Classes (HCCs), that can be used clinically and for research purposes to preoperatively classify hysterectomies depending on the risk of complications.

Johanna Wagenius, Andrea Stuart, and Jan Baekelandt declare consultancy for Applied Medical. Michael Conditt declares consultancy for Momentis Medical. Karin Källen and Hanna Källen Murray have no conflict of interests.

The study was supported by grants from Stig and Ragna Gorthon foundation. The funders had no role in the study design, data analysis, data interpretation, or writing the report.

This study protocol was reviewed and approved by the Swedish National Ethical Committee, approval date 220413, Reference No. 2022-01489-01. Data in this study were obtained from the Swedish National Quality Registry GynOp. Participants did not provide study-specific written informed consent; however, all individuals had previously consented to inclusion in GynOp and to the use of their data for research and quality improvement according to Swedish registry regulations. In Sweden, written study-specific informed consent is not required when using data from a national quality registry. Participants have already received standardized information and given informed approval to be included in GynOp, with full rights to decline or withdraw. The Swedish national Ethics Committee therefore does not require additional written consent for registry-based research.

All authors contributed to the study conception and design. Material preparation and data collection and analysis were performed by Andrea Stuart, Michael Conditt, Karin Källen Hanna Källen Murray, and Jan Baekelandt. The first draft of the manuscript was written by Andrea Sturt and all authors commented on previous versions of the manuscript and read and approved the final manuscript.

We here present a novel comprehensive prediction model for the risk of any complication after minimally invasive hysterectomy. This model is predictive rather than explanatory as it was developed in one dataset and validated in an independent dataset. To facilitate clinical use, a free online calculator (Complication Risk Calculator – iNOTESs) is available for estimating individual complication risk preoperatively. We used a strict definition of complication, including any factor that deviates from a successful uneventful hysterectomy. The main risk factors for complications after hysterectomy were uterus size, which interacted with surgical technique, and also indication for hysterectomy, BMI, previous CS, and age. The presence of smoking, the need of interpreter, use of drugs for cardiovascular, diabetes, psychiatric disease had no impact on the risk of complications. Upon validation, our prediction model showed high sensitivity to accurately predict the risk of composite complication. Conversion to open surgery was included in the composite outcome because it represents a deviation from the intended minimally invasive approach and is associated with increased morbidity. Although the model showed only modest discriminative ability at the individual level, this is expected in surgical prediction models where outcomes are influenced by intraoperative factors not captured in registries. Importantly, the model demonstrated strong calibration, with close agreement between predicted and observed risks across strata. For clinical counseling, and research stratification, accurate estimation of group-level risk is more relevant than perfect individual-level prediction, and the model performs well in this regard. A key strength of this model is the explicit incorporation of the interaction between uterus size and surgical technique. This relationship is well recognized clinically but has not been quantified in previous prediction models. By integrating this interaction, the model reflects real-world surgical complexity more accurately than models based solely on main effects. A strength of our study was the large comprehensive cohort studied, with over 60,000 hysterectomies, with data from four different national registries, not only identifying surgical variables but also sociodemographic factors and factors that might impact the risk of complications such as prescriptions of antipsychotics or immunosuppressive drugs. The nationwide design minimizes selection bias and reflects real-world practice across hospitals of varying size and surgical expertise. This enhances the external validity of the model and increases its applicability to diverse clinical settings. A weakness of our study was that surgeon expertise could not be included in the model. Experienced surgeons generally have better outcomes in performing hysterectomies, particularly for simple cases, exhibiting shorter operation times and lower complication rates compared to their less experienced counterparts. Furthermore, they are often tasked with more complex cases, which, while carrying a higher risk of complications and extended surgical durations, benefit from the surgeon’s greater skill and knowledge, potentially mitigating some of these risks. Medication-based comorbidity measures were only available from 2005, which may have led to under-ascertainment of comorbidities in earlier years, which is also a weakness of our study. A difficulty regarding analysis of mode of hysterectomy as a predictor of outcome is the difference in surgeon volume in Sweden. In 2023, the same number of hysterectomies were performed laparoscopically and robotically, but there were 162 laparoscopic surgeons and 100 robotic surgeons, leading to higher surgical volume per robotic surgeon. A previous publication from Sweden, using the same GynOp Registry, showed a difference in surgeon and hospital volume, with a lower caseload per year in the LH than in the RH group [ 14 ]. Pepin et al. [ 15 ] published a prediction model for complications after LH, with data from the USA, with a composite complication rate of 14.1%. They included ASA class in their model, in contrast to our model. In Sweden, 98% of the benign hysterectomies are performed on patients with ASA 1–2 [ 16 ]. By including detailed prescriptions of cardiovascular, diabetic, or psychiatric drugs as indicators of preoperative health, we attempted to analyze more in detail if there was any preexisting medical diagnosis that increased the risk of complications. Another difference from the prediction model by Pepin et al. [ 15 ] is that healthcare is organized differently in the USA and Sweden, the latter in which healthcare is socialized and in almost all cases funded by the government. In contrast to previously published models, including a Canadian model from 2022 et al. [ 17 ], our approach incorporates three MIS techniques, integrates the interaction between uterus size and surgical technique, and uses four national registries including socioeconomic and pharmacological data. Our comprehensive model includes data from four different registries, including socioeconomic, educational, medical, pharmacological, and surgical data. Furthermore, our model includes three different MIS techniques and takes the interaction of technique and uterus weight into account. We used a stricter cutoff of the composite for complication; those without a complication had a straightforward hysterectomy with no deviation from the intended intra- and postoperative route. Therefore, we included factors such as data from the mandatory Drug Registry, including all patients receiving postoperative antibiotics within 6 weeks in our model, showing a rate between 10 and 15% depending on technique. The presence of a prescription of postoperative opioids was not included in the composite complication, as we interpreted it mainly to mirror local clinical tradition, rather than postoperative pain. A prescription of opioids was more common regardless of complications among patients with robotic and laparoscopic hysterectomies compared to VH. Settnes et al. [ 7 ] showed in a Danish national cohort study including 50,000 benign hysterectomies an approximately 50% risk reduction of both minor and major complications if VH was performed due to prolapse versus VH for non-prolapse, OR 0.46 (0.39–0.55), and approximately doubled increased risk of major complications if the indication for hysterectomy was endometriosis, OR 1.96 (1.58–2.42). The same study showed that the risk of complications was approximately 60% increased among women under 55 years old compared to over 55 years old. The results and risk estimates are in analogy with our results from the Swedish databases. A large meta-analysis [ 5 ] including approximately 55,000 women showed that women with a previous CS have increased risk of urinary tract injury (adjusted OR = 2.21, 95% CI = 1.46–3.34), gastrointestinal tract injury (adjusted OR = 1.83, 95% CI = 1.11–3.03), postoperative infections (OR = 1.44, 95% CI = 1.22–1.71), wound complications (pooled unadjusted OR = 2.24), and reoperation (unadjusted OR = 1.46, 95% CI = 1.19–1.78). Also, these results are in analogy with our results, showing approximately 20% increased risk of complications after CS. Previous cesarean section is a known risk factor for hysterectomy complications, showing increased risk of organ injury and postoperative infection with 74% and 26%, respectively [ 4 ]. Our results showed an increased risk of the composite complication with 19% after CS compared to parous women without CS. Socioeconomic factors such as race are associated with both hysterectomy technique used and complications. Black women in the USA are half as likely to have a minimally invasive hysterectomy (LH or VH) compared to white women [ 8 ]. Smaller hospitals, black and older patients are more prone to undergo an open abdominal hysterectomy versus an LH/VH, and the rate of perioperative complications was affected by patient race and financial status (Medicaid vs. no Medicaid) [ 9 ]. Our results showed that older women are more likely to have open surgery instead of MIS, and the risk of complications increased with 30% if the patient’s country of birth was outside of Europe. A Danish study including 22,000 benign hysterectomies [ 10 ] showed that the risk of complications increased with decreasing socioeconomic position. Women with lower educational level and unemployed women had higher odds of infection, bleeding, reoperation, and readmission than women with more than high school education and employed women. Unemployed women had higher odds of hospitalization >4 days than women in employment. In contrast to Denmark, in Sweden we showed a decreased risk of adverse outcome among women with the lowest educational level. Previous research has shown a lower risk of sphincter tears in Sweden among women with lower educational level [ 18 ]. The results are surprising and suggest that disparities in healthcare access and diagnostic practices may lead to women with higher educational levels being more frequently diagnosed with complications, rather than experiencing them at a higher rate. As we do not believe that high educational level is a true risk factor for complications due to hysterectomy, and a factor that cannot be taken into account when counselling the patient for surgery, we decided not to include educational level in the multiple model. We present a novel assessment tool for the risk of hysterectomy complications based on established risk factors. This tool converts data that a surgeon may already intuitively understand into a tangible percentage that can be compared and analyzed. In a second step, we propose a new classification system, HCCs 1–5, based on the results from the prediction model. Proposed HCCs 1–5 offer a structured classification system that aids clinicians in making informed decisions based on predicted outcomes and complications associated with hysterectomy procedures. By categorizing complications into distinct classes, this system not only enhances the clarity of communication among healthcare professionals but also serves as a valuable tool for research initiatives aimed at improving patient safety and surgical practices.