1. Study design
After obtaining approval from the Institutional Review Board of Dankook Dental University Hospital (No. DKUDH IRB 2020-09-007), records of patients admitted to the Oral and Maxillofacial Surgery (OMFS) Department, Dankook University Hospital, for the treatment of maxillofacial infections from January 2011 through September 2020 were reviewed. The billing department performed a computer search of billing data attached to the electronic medical record numbers of patients with the relevant diagnostic codes from the seventh revision of the Korea Informative Classification of Diseases (KCD7). KCD7 codes used to identify relevant patients were K04.6 (periapical abscess with sinus), K04.7 (periapical abscess without sinus), K05.21 (periodontal abscess of gingival origin with sinus), K11.3 (abscess of the salivary gland), K12.2 (cellulitis and abscess of the mouth), K14.0 (abscess and ulceration of the tongue), J36 (peritonsillar abscess), J39.0 (retropharyngeal and parapharyngeal abscess), and J39.1 (other abscess of the pharynx). Retrieved positive records were then cross-checked with the relevant codes to ensure that admissions were related to a maxillofacial infection. An additional, complete manual review of patient admission data was also performed to identify additional records not identified by the electronic search process. Exclusion criteria for this study were patients treated only in an outpatient setting, incomplete data acquisition from electronic charts, patients not yet discharged from the hospital, patients who were initially admitted for reasons other than maxillofacial infections but who developed infection secondary to other maxillofacial surgeries, and patients for whom the OMFS service was not consulted. To focus on the characteristics and outcomes of more severe, potentially fatal infections, the current study was limited to cases that required hospital admission for treatment. To ensure that the exclusion criteria were applied and that recording of the relevant data was accurate, each chart was reviewed twice by two separate reviewers. At the reviewing stage, the reviewers were blinded to the hospital bills of each subject to eliminate the possibility of bias in collecting data. Names of the subjects were not revealed to protect personal information.
2. Independent variables
All variables extracted from patient records were recorded and organized on an electronic Microsoft Excel 2021 (Microsoft, Redmond, WA, USA) spreadsheet. Independent variables examined in the present study comprised a heterogeneous set of demographic, individual health status, hospital course, infection, and treatment-related variables.
Demographic variables were age, sex, ethnicity (Korean or not Korean), insurance coverage status (National Health Insurance [NHI], medical aid, or uninsured), and admission year. Age, sex, and ethnicity were determined based upon information in patients’ medical records. Information regarding insurance status was provided by the billing department. Admission year was determined based on admission and discharge dates. When the patient was admitted at the end of one year and discharged the next year, the admission year was considered the year with the longer length of stay (LOS).
Variables related to an individual’s health status were the presence of any kind of comorbidity, the number of comorbidities, smoking, and the burden of each type of comorbid condition. All these variables were noted as written in the nursing admission assessment form. To assess the burden of comorbid conditions such as diabetes, endocrine disease, nephrotic disease, hepatic disease, cardiovascular disease, and psychological disease, each condition was tested bivariately and only those associated significantly with outcomes were included in the final statistical analysis.
Hospital course-related variables were route of admission (via outpatient clinic or emergency room [ER]), hospital arrival time (morning [00:00 to 08:00], day [08:00 to 16:00], or evening [16:00 to 24:00]), and time elapsed from symptom onset to admission, from admission to first surgery, and from symptom onset to first surgery (in days). Route of admission and hospital arrival time were determined from information provided by the billing department. Date of symptom onset was recorded as described by the patients at the time of the first ER or outpatient clinic visit. Date of the first surgery was determined from the procedure and surgery notes.
Infection-related variables included the infection’s etiology (odontogenic or non-odontogenic), number of infected anatomical spaces, infected location, presence of deep neck space infection, presence of necrotizing fasciitis, Flynn score, C-reactive protein (CRP) level, white blood cell (WBC) count, glucose level upon admission, body temperature upon admission, and peak body temperature during hospital stay. Information regarding the infection’s etiology and spaces involved was determined by concomitant review of discharge summaries and enhanced neck computed tomography scans. An odontogenic infection severity score was assigned to each infection based on a scoring system suggested in a previous study by Flynn et al.
4
, in which each infected fascial space is scored on the basis of its proximity to the airway, other vital structures, and likelihood of preventing airway access; the total score is the aggregate of these values. Thus the vestibular, subperiosteal, infraorbital, canine, and buccal spaces were assigned a score of 1, the submandibular, submental, sublingual, and masticatory spaces a score of 2, and the lateral pharyngeal, retropharyngeal, pretracheal, mediastinal, and intracranial spaces a score of 3. Body temperatures were determined through review of the patient’s daily clinical examination form, and were categorized as 37.5℃ or more or less than 37.5℃.
Treatment-related variables included antibiotic regimen, performance of incision and drainage, tracheostomy, any type of surgical intervention after admission, general anesthesia (GA), additional surgery, additional operating room (OR) uses, total number of OR visits, surgeries, type and location of surgeries received, ICU use, and length of ICU stay. Antibiotic regimen was categorized based upon the most commonly utilized combinations of antibiotics during the reviewed period, and was clarified by reviewing the daily prescription data. All information regarding surgical interventions and GA was obtained by comprehensive review of the information recorded on emergency department records, daily procedure notes, daily inpatient notes, surgery notes, anesthesia notes, and discharge summaries. Among surgery-related variables, length of surgery was calculated as the difference between the recorded ‘procedure start’ and ‘procedure end’ times on the anesthesia note. Information regarding ICU stay was provided by the billing department.
4. Statistical analysis
All retrieved data were compiled, matched with the outcome data, and imported into IBM SPSS Statistics (ver. 27.0; IBM, Armonk, NY, USA) for statistical analyses. Descriptive statistics were computed for all variables. Because data regarding hospital charges were highly skewed and not normally distributed, the billing data was log-transformed to adjust for violation of the normality assumption. For preliminary univariable analysis, data were analyzed using the t-test, Mann–Whitney test, analysis of variance, Kruskal–Wallis test, Pearson’s correlation analysis, χ2 test, and Fisher’s exact test as appropriate for the bivariate analysis performed. A P-value less than 0.05 was considered to be statistically significant. All variables with statistically significant associations with the dependent variables (P<0.05) from the initial univariable analysis were used to formulate a preliminary multiple linear regression model for the continuous dependent variables, and a preliminary multiple logistic regression model for the binary dependent variables. The final multiple regression models included all variables from the preliminary model with P-values lower than 0.05. In all regression models, adjustments were made for the clustering of outcomes. All statistical analyses were two-sided.