Effects of preoperative oral management by dentists on postoperative outcomes following esophagectomy: Multilevel propensity score matching and weighting analyses using the Japanese inpatient database.

The purpose of this study was to investigate the effects of preoperative oral management (POM) by dentists on the incidence of postoperative pulmonary complications (PPCs), length of hospital stay, medical costs, and days of antibiotics administration following both open and thoracoscopic esophagectomy.Dental plaque is an established risk factor for postoperative pneumonia, which could be reduced by POM. However, few clinical guidelines for cancer treatment, including those for esophageal cancer, recommend POM as routine perioperative care.We extracted data of esophagectomy cases from the Japanese Diagnosis Procedure Combination database. We subsequently conducted propensity score (PS) analyses for multilevel data, including matching, inverse probability of treatment weighting (IPTW), and standardized mortality ratio weighting (SMRW), to estimate the effect of POM by dentists on the outcomes of esophagectomy.We analyzed 3412 esophagectomy cases of which 812 were open, and 2600 were thoracoscopic surgery. In IPTW analysis to estimate the average treatment effect, the risk difference of postoperative aspiration pneumonia ranged from -2.49% to -2.02% between the POM and control groups of both open and thoracoscopic esophagectomy cases. IPTW analyses indicated that the total medical costs of thoracoscopic esophagectomy were reduced by 221,200 to 253,100 Japanese Yen (equivalent to about $2000-$2200). In PS matching and SMRW analyses to estimate average treatment effect on treated, there was no difference in outcomes between the POM and control groups.Our results suggested that in patients undergoing open or thoracoscopic esophagectomy, POM by dentists prevented the occurrence of postoperative aspiration pneumonia. It could also reduce the total medical costs of thoracoscopic esophagectomy. Thus, POM by dentists can be considered as a routine perioperative care for all patients undergoing esophagectomy, regardless of the expected risk for PPC.

Introduction

Esophageal cancer ranked ninth (11th in Japan) in the number of incident cases and sixth (7th in Japan) in the number of deaths globally, as of 2013.[ Esophagectomy, an essential treatment modality for esophageal cancer, is a highly invasive procedure with a high risk of postoperative complications,[ especially pulmonary complication as the independent predictor of poor long-term survival following esophagectomy.[

Postoperative pulmonary complications (PPCs) increase the mortality, health-care costs, and length of hospital stay (LOS).[ Dental plaque, a biofilm that serves as a reservoir of oral bacteria, is a known risk factor for postoperative pneumonia including aspiration pneumonia and ventilator-associated pneumonia,[ thus perioperative oral management could reduce the occurrence of postoperative pneumonia. In Japan, the treatment fees for perioperative oral management by dentists were newly included in the dental fee schedule of the National Health Insurance to prevent postoperative complications, such as aspiration pneumonia, in 2012.[ Furthermore, additional fees for surgery preceded by preoperative oral management (POM) which can be claimed when conducted within 1 month prior to cancer surgery of the orofacial, neck, thoracic and abdominal areas, and cardiac surgery were imposed on the medical fee schedule in 2014 as incentives to promote POM.[ However, few clinical guidelines for cancer treatment recommend oral management by dentists as routine perioperative care. Thus, the purpose of this study was to investigate the effects of POM by dentists on various postoperative outcomes, including PPCs, postoperative LOS, medical costs, and days of antibiotics administration following open and thoracoscopic esophagectomy.

Methods

Data source

We analyzed the Diagnosis Procedure Combination (DPC) data of esophagectomy cases from the database of the DPC research group,[ which is funded by the Ministry of Health, Labour and Welfare (MHLW), Japan. The DPC/per-diem payment system (PDPS) is a Japanese prospective payment system applied to acute-care hospitals. DPC data comprise claims of inpatients, as well as patients’ information of discharge summaries, such as height, weight, activities of daily living (ADL) indices, and International Classification of Diseases, Tenth Revision (ICD-10) codes, which are classified as the main diagnosis, cause of admission, most and the 2nd most medical resource-intensive diagnoses, up to 10 comorbidities, and 10 complications; in addition, it contains diagnosis-specific information, such as the TNM classification for cancer cases.

Inclusion and exclusion criteria

We included esophagectomy cases discharged between April 1, 2016, and March 31, 2017 (fiscal year 2016). We selected the surgical codes for open and thoracoscopic esophagectomy of the cervical, thoracic, and abdominal esophagus with ICD-10 codes of esophageal cancer (C15.x) as the cause of each admission. We included new esophageal cancer cases, not recurrent cases, in which the first surgery under general anesthesia during each admission period was esophagectomy. We excluded emergency and unexpected admissions, because the additional fees for surgery preceded by POM can be claimed only for those cases with POM administered within 1 month before the day of surgery. We further excluded cases with inconsistent data, such as having both codes of open and thoracoscopic esophagectomies, which were not considered as valid claims. To deal with missing values, we adopted complete case analysis (CCA) based on a recent simulation study which suggested that CCA could reduce the bias in propensity score (PS) matching and weighting.[ For CCA, cases with missing values of the ADL indices, body mass index (BMI), and/or prescription information of antibiotics were excluded.

Outcomes of interests

In our study, the outcomes included the incidence of PPCs, postoperative LOS, medical costs, and days of antibiotics administration after esophagectomy. The cases with PPCs were defined as those with ICD-10 codes of pneumonia due to infectious organisms (J13.x–J18.x), aspiration pneumonia (J69.x), and/or postprocedural respiratory disorders (J95.8 and/or J95.9) as complications. We presented the incidence of aspiration pneumonia along with that of all PPCs. We calculated fee-for-service (FFS) costs as the medical costs, not the claimed costs by PDPS, to signify costs as the actual amount of consumed medical resources. The days of antibiotics administration was selected as a proxy of the days needed to treat postoperative infection.

Statistical analyses

We assigned cases with claims of additional fees for surgery preceded by POM as the POM group, and cases without these claims as the control group. We then performed PS matching and weighting analyses for clustered data, since the DPC data was considered to have a clustered structure of patients nested in each hospital.

The covariates for estimating PSs included the sex, age strata (<40, 40–49, 50–59, 60–69, 70–79, and >80 years old), BMI (<18.5 and ≥25), any dependency on ADL (Barthel index <100), smoking history (Brinkman index of 0, 1–399, 400–799, 800–1199, and ≥1200), TNM classification (T≥3, N≥1, and M = 1), and preoperative respiratory rehabilitation during the same hospitalization period as that for surgery. The covariates of comorbidities included 32 categories suggested by Gagne et al,[ which are the combined Charlson and Elixhauser comorbidities. We identified each comorbidity using ICD-10 coding algorithms described by Quan et al.[ Using these covariates, logistic regression models were fitted to estimate single-level PSs. Fixed-effects PSs were estimated by fitting logistic regression models with the listed covariates and dummy variables of each hospital as intercepts, and random-effects PSs were estimated by fitting generalized linear mixed models with the covariates as fixed-effects and hospital codes as random-effects.[ The dependent variables of 3 PS models were the indicator variables of the claims for additional fees of surgery preceded by POM.

We conducted 4 matchings to estimate the average treatment effect on treated (ATT) using the estimated PSs: single-level, preferential within-cluster, fixed-effects, and random-effects matching; the latter 3 are matching for clustered data described by Arpino and Cannas.[ For single-level matching, which does not consider the clustered data structure, cases in the POM group were matched within all cases in the control group using single-level PSs. For preferential within-cluster matching, cases in the POM group were matched only within cases at the same hospital as that of the POM group using single-level PSs. This step might result in loss of many cases, so additional matching for unmatched cases in the POM group were conducted; cases that could not be matched within the same hospitals were matched with cases at hospitals other than the nested hospitals, using single-level PSs. For fixed-effects and random-effects matching, the fixed-effects PSs and random-effects PSs were used for matchings, respectively, where cases in the POM group were matched within all cases in the control group. For all matching, we performed one-to-one nearest-neighbor matching with replacement using a caliper of 0.2, as recommended by Austin.[ In summary, the single-level matching ignored the clustered structure of the data, preferential within-cluster matching considered it by means of the matching method, and fixed-effects and random-effects matching considered it through estimation of PSs.

To estimate average treatment effect (ATE) and ATT by PS weighting, we used inverse probability of treatment weighting (IPTW) method and standardized mortality ratio weighting (SMRW) method,[ respectively. For PS weighting analyses, as in matching analyses, considering the clustered structure at either the stage of estimating the PSs or treatment effects improved the quality of estimates.[ Thus, we calculated the weights using 3 PSs estimated for matching. Using these weights, we estimated the ATE and ATT as risk differences, and differences in days or Japanese Yen (JPY).

The balances of covariates between the POM and control groups before and after matching and weighting were checked by standardized differences.[ Charlson comorbidity index and Elixhauser/van Walraven comorbidity index[ were presented instead of the 32 comorbidities to simplify the comparisons. To estimate P values between the groups, Student t test (or Welch–Satterthwaite t test for covariates with unequal variances) and Fisher exact test (if not applicable, χ2 tests) were performed for baseline populations. After matching analyses, paired t test and McNemar test were performed. After weighting analyses, generalized linear models were fitted to estimate P values and confidence intervals using robust variances.[

SAS software version 9.4 (SAS Institute Inc, Cary, NC) was used for all analyses: PROC LOGISTICS and PROC GLIMMIX were used to estimate PSs, PROC PSMATCH for matchings, and PROC GENMOD for weighting analyses. Two-tailed significance level of 0.05 was used for all tests.

Sensitivity analyses

We used PS analyses methods for clustered data with unknown cluster-level confounders. However, we could calculate 1 cluster-level confounder of the hospital volume. Based on the annual cases at each hospital, we divided the hospitals into 3 categories of hospital volume: 1 to 11, 12 to 23, 24 or more cases per year. As the first sensitivity analysis, we incorporated these variables into the 3 PS models, and then conducted matching and weighting analyses.

Among several methods that deal with missing variables in PS analyses, the missing indicators method may reduce the bias.[ As the second sensitivity analysis, we added the missing indicators of smoking index, BMI, and ADL scores into our PS models.

The third sensitivity analysis is a solution for the drawback of the PS weighting analyses. In PS weighting, the weight can be very big or small when the PS is near 0 or 1. Therefore, we used trimmed (or truncated) weights for the third sensitivity analysis. From among several ways to trim weights, we adopted the method using 1st and 99th percentiles of estimated weights.[ The weights smaller than the 1st percentile of all weights were trimmed to the same values as the 1st percentile, and the weights bigger than the 99th percentile were trimmed to the same values as the 99th percentile.

Ethical considerations

This study was conducted in accordance with the Ethical Guidelines for Medical and Health Research involving Human Subjects of the MHLW, Japan. The Ethics Committee, Graduate School of Medicine Kyoto University approved the study (approval number: R0135).

Results

Study population

Figure 1 shows the flow of case selection. We extracted 3412 esophagectomy cases and excluded 812 cases; thus, 2600 cases were included in the final analysis. The number of hospitals was 202 for open surgery, and 220 for thoracoscopic surgery.

Figure 1

The flow of case selection. POM = preoperative oral management by dentists.

Patients’ characteristics at baseline

Table 1 shows the comparison of patients’ characteristics between the POM groups and control groups at baseline. The complete lists of comorbidities are presented in Supplemental Digital Content 1. At baseline, the proportion of patients with preoperative respiration rehabilitation was higher in the POM groups for both surgery types. The proportion of patients with Barthel indices of < 100 was higher in the control group of thoracoscopic surgery than that of the POM group.

Table 1

Comparison of the baseline characteristics between the preoperative oral management groups and control groups of open and thoracoscopic esophagectomy.

Balance of covariates of the matched and weighted populations

Tables 2 and 3 show the standardized differences of the POM groups and control groups before and after matchings, and after weightings, respectively. For both surgery types, the mean absolute standardized differences of the PS weighting analyses were smaller than those of the PS matching analyses, which indicated that the 2 groups were balanced better in weightings than in matchings. Among the mean absolute standardized differences after matching, that of the random-effects matching was smallest for open esophagectomy, whereas that of the fixed-effects matching was smallest for thoracoscopic esophagectomy.

Table 2

Standardized differences of the selected covariates of the preoperative oral management groups and control groups at baseline and after matching.

Table 3

Standardized differences of the selected covariates of the preoperative oral management groups and control groups after weighting analyses.

Outcomes

Table 4 shows the postoperative outcomes of the POM and control groups at baseline, and after matching. At baseline, the incidence of PPCs was higher in the control group of open esophagectomy (16.4% vs 10.6%), and the medical costs were lower in the POM groups of thoracoscopic surgery (3.4 vs 3.7 million Japanese Yen). Matching analyses revealed that none of the outcomes of the POM groups were better significantly than those of the control group except for all PPCs after open esophagectomy through fixed-effects matching. In contrast, the incidence of postoperative aspiration pneumonia was consistently low by 2.0% to 2.5% in the POM groups through all IPTW analyses which estimated ATE, except for the analysis of open esophagectomy population using weights derived from single-level PSs (Table 5). The medical costs of thoracoscopic esophagectomy were significantly low by 221,200 to 253,100 Japanese Yen in the POM groups through the IPTW analyses (Table 5).

Table 4

Comparison of the postoperative outcomes between the preoperative oral management groups and control groups at baseline and after matching.

Table 5

Comparison of the postoperative outcomes between the preoperative oral management groups and control groups in IPTW analyses.

Supplemental Digital Contents 3 and 4 show results of the sensitivity analyses of matching and weighting analyses, respectively. In the matching analyses of thoracoscopic esophagectomy, all outcomes showed no statistical significance except for the incidence of aspiration pneumonia through single-level matching adding missing indicators. For open esophagectomy, however, the postoperative LOS was longer in the POM groups through 3 matching analyses (37.6 vs 32.7 in single-level matching using missing indicator; 37.3 vs 31.7 in preferential within-cluster matching, and 36.8 vs 31.3 in fixed-effects matching using variables of hospital volume), whereas the incidence of PPCs was lower in the POM groups through 2 matching analyses (9.8% vs 18.0% in single-level matching and 9.8% vs 15.6% in preferential within-cluster matching using missing indicator). In the SMRW analyses, the outcomes showed no significant change. However, through the IPTW analyses using the trimmed weights derived from multilevel PSs and weights from the random-effects model with missing indicators, the P values for the incidence of aspiration pneumonia following thoracoscopic esophagectomy rose above .05 (relative differences and P values of the incidence of aspiration pneumonia using fixed-effects PS and random-effects PS in main analysis were −2.1%, P = .031 and −2.0%, P = .039, respectively; in analysis using trimmed weight, they were changed to −1.9%, P = .060, −1.9%, P = .054).

Discussion

We conducted multilevel PS analyses using the Japanese DPC data as clustered data to investigate the effects of POM by dentists on postoperative outcomes of esophagectomy. In the IPTW analyses, POM by dentists was related to the prevention of postoperative aspiration pneumonia for both open and thoracoscopic esophagectomy, and also related to the reduction of medical costs of thoracoscopic esophagectomy. The PS matchings and SMRW analyses indicated that POM by dentists had no effectiveness to improve the postoperative outcomes.

We estimated both the ATT and ATE of POM by dentists and showed that the risk reduction of postoperative aspiration pneumonia achieved in terms of ATE. These results suggested that POM by dentists for all esophagectomy patients, not just patients with expected high risk of PPC based on clinical factors that were not included in the DPC data, could be beneficial to prevent postoperative aspiration pneumonia. Although the preoperative oral management would cost 8000 to 16,000 JPY per person which is equivalent to around $70 to $140 based on the fee schedule as of 2016, these costs would be offset based on the IPTW analyses results of further reduction in the medical costs (221,200 to 253,100 JPY, equivalent to about $2000 to $2200, for thoracoscopic esophagectomy). The 2012 Japanese guidelines of esophageal carcinoma recommended preoperative oral care before esophagectomy.[ However, this recommendation has been deleted in the newest guidelines published in 2018,[ possibly since the levels of evidence supporting the effectiveness of POM on postoperative complications were not high enough.[ Risk factors of PPC with good evidence including congestive heart failure, ASA class of 2 or more, advanced age, chronic obstructive pulmonary disease, functional dependence, prolonged surgery, and lower preoperative serum albumin level were reported.[ The relationship between POM by dentists and PPC following esophagectomy is not supported by reports with high-level evidence. However, recently, a multicenter study (n = 539) demonstrated that POM by dentists could reduce postoperative pneumonia following esophageal cancer surgery,[ and a study using a large-scale national administrative claims database (n = 16,177 for esophageal cancer, among n = 509,179 for 6 cancers) showed that POM by dentists reduced PPCs of major cancer surgery.[ Nevertheless, the current guidelines of esophagectomy including Japanese guidelines of esophageal carcinoma[ and enhanced recovery for esophagectomy,[ do not recommend POM by dentists as routine perioperative care. In our study, we showed that POM by dentists could prevent postoperative aspiration pneumonia following both open and thoracoscopic esophagectomy, and save the medical costs of thoracoscopic esophagectomy based on IPTW analyses which estimated ATE. These results suggested that POM by dentists should be recommended as routine perioperative care for all patients undergoing esophagectomy, regardless of the surgery type.

To analyze data with clustered structure, multilevel regression models are used frequently. Despite the increasing use of PS analyses in medical and healthcare studies, few studies have performed PS analyses considering clustered data structure.[ With regard to both PS matching and weighting analyses, the results of simulation studies showed that by ignoring clustered structures there was biased estimation of the treatment effects.[ To the best of our knowledge, our study is the first PS analyses of Japanese DPC data as clustered data from multiple hospitals; so far, only a few studies incorporated hospital volume as a variable in their PS models. We conducted PS analyses with and without considering the clustered structure and showed that the results obtained using the 2 approaches could be different. An earlier report indicated that hospital volume, one of the cluster-level confounders, influenced mortality rates following esophagectomy in Japan,[ as in other countries. In our study, PS weighting analyses revealed that by incorporating the variables representing hospital volume into the PS models, the estimates of analyses using single-level PSs fluctuated more than those of analyses using multilevel PSs (Supplemental Digital Content 4), which implies that analyses that do not consider the clustered structure of data are not robust enough, consistent with the simulation studies. Moreover, hospital volume is not the only cluster-level confounder; the institutional heterogeneity of esophagectomy was previously reported to persist even among high-volume hospitals in Japan.[ Researchers should consider the unmeasured cluster-level confounders in PS analyses of data with clustered structure.

Among previous studies to determine the risk factors of complications following esophagectomy in the Japanese population, some studies used databases,[ such as the National Clinical Database (NCD) and the National Database of Health Insurance Claims and Specific Health Checkups of Japan (NDB).[ The NCD contains variables that are almost the same as those of the American College of Surgeons’ National Surgical Quality Improvement Program,[ and the NDB contains all electric health insurance claims of Japan. The NCD has various clinical variables such as the values of preoperative laboratory tests, and intraoperative blood loss, which are not included in the claims database. The DPC used in our study comprises a mixture of claims and discharge summaries. Since these databases cannot be linked currently, the DPC is the most popular database for various healthcare studies due to its mixed nature. The DPC data contains some diagnosis-specific variables such as the TNM classification for cancer patients, and information of both PDPS and FFS-based medical services at admission. Using the DPC data to investigate the effects of POM by dentists, we determined multiple outcomes including the incidence of PPCs and the total medical costs, with adjusting for various patient-level confounders.

Our study has some limitations. First, the DPC data does not contains clinical data except for some diagnosis-specific variables. Some of the known risk factors, such as the results of laboratory tests, are not available, which may cause insufficient adjustment of the risk factors for various outcomes. Doctors may consider referring patients with high risk of PPC based on results of preoperative laboratory tests to a dentist for POM. However, this would lead to underestimating the effects of POM,[ so the result of reduced rate of postoperative aspiration pneumonia in our study should be consistent. Second, we only had information of each admission period, but not that of before and after admission. We assigned cases with claims of additional fees for surgery preceded by POM as the POM group, not with claims of treatment fees for perioperative oral management by dentists. It is possible that the additional fees were not input by the hospital at which surgery was performed, nevertheless POM had been administered by dentists before surgery. However, this would also lead to underestimating the effects of POM, if some cases with POM accidentally assigned to the control group. Preoperative chemotherapy, which is regarded as standard procedure for resectable esophageal cancer cases in Japan,[ could not be reviewed in our study if conducted before admission for surgery. Nishino et al[ reported that oral care may be provided to patients who received chemotherapy since oral mucositis is a frequent adverse effect following chemotherapy, which could reduce the incidence of postoperative pneumonia. However, a previous multicenter study incorporating various risk factors including POM by dentists and neoadjuvant chemotherapy reported that neoadjuvant chemotherapy was not related to the incidence of postoperative pneumonia.[ We were unable to conduct patient follow-up after discharge, and so, could not determine the LOS in patients with transfer from the hospital at which esophagectomy was performed to another hospital after discharge. The DPC data has the variable of destination after discharge including the home, another hospital, and the nursing home. We were unable to determine the overall LOS if a patient did not return home after esophagectomy, but could check the balance of this variable instead. The absolute standardized differences of the 3 postmatching populations and 2 weighted populations were > 0.1; more patients in the POM groups through fixed- and random-effects matching, and SMRW analyses using fixed- and random-effects PSs of open esophagectomy were transferred to other hospitals, whereas less patients in the POM groups through random-effects matching of thoracoscopic esophagectomy were transferred. The absolute standardized differences in all of the IPTW analyses were ≤0.1, suggesting that the effects of transfer were minimal.

Conclusion

Our results suggested that POM by dentists for all patients undergoing open or thoracoscopic esophagectomy could reduce the incidence of postoperative aspiration pneumonia. It also could reduce the total medical costs of thoracoscopic esophagectomy. Thus, POM by dentists can be considered as routine perioperative care for all patients undergoing esophagectomy, regardless of the expected risk for PPC.

Author contributions

Conceptualization: Jung-ho Shin, Susumu Kunisawa, Yuichi Imanaka.

Data curation: Kiyohide Fushimi.

Formal analysis: Jung-ho Shin.

Funding acquisition: Yuichi Imanaka.

Methodology: Jung-ho Shin, Yuichi Imanaka.

Project administration: Susumu Kunisawa.

Resources: Susumu Kunisawa, Kiyohide Fushimi.

Supervision: Yuichi Imanaka.

Writing – original draft: Jung-ho Shin.

Writing – review & editing: Susumu Kunisawa, Kiyohide Fushimi, Yuichi Imanaka.