This retrospective cohort study was conducted at Oulu University Hospital, Finland. The study protocol was accepted by the hospital administration (ref 42/2021). According to Finnish legislation, patient consent is not required for a retrospective registry study. Following the local policy, no statement from the ethics committee was obtained due to the retrospective study design. All patients who underwent tumour resection and neck dissection (ND) with or without free-flap surgery (FFS) due to HNC between 1 January 2014 and 31 December 2019 were included in this study. Patients had cancer of the lip, oral cavity, pharynx, larynx, nasal cavity, paranasal sinus, or salivary glands, or thyroid cancer, but the focus of the study was on major head and neck surgery.

Patient data, including age, gender, American Society of Anesthesiologists (ASA) classification, tumour site, and medical and surgical complications was collected from the medical records. The recorded medical complications included pneumonia, acute myocardial infarction, pulmonary embolism, deep venous thrombosis, stroke, arrhythmia, acute kidney injury, and delirium. The recorded surgical complications included postoperative bleeding, surgical site infection, free flap loss, chyle leak, and need for reoperation. The same classification has been used previously (Lahtinen et al. 2018). The patients were followed until 31 December 2021, and dates of death were obtained from the hospital’s patient data management system, which is regularly updated from the Finnish death registry.

This study was conducted at Oulu University Hospital, which is one of five cancer centres in Finland and provides operative and advanced oncological care for the population living in Northern Finland. For patients suffering from HNC, both ablative and reconstructive surgery are available in the unit. The surface area of the hospital district is 172,900 km2, and the number of inhabitants in 2018 was 738,700. In this area, the distance to the nearest primary health care centre can be up to 100 km, and the journey from the northernmost municipality to the nearest central hospital takes several hours, even longer to Oulu University Hospital.

The postal codes of the residential areas were used to classify patients as urban and rural populations and to determine the SES. Statistics Finland provides open population data for each postal code area in Finland (Paavo postinumerotietokanta). The patient data were combined with the data for the postal code areas, and 2018 was used as the index year. The patients were from 142 postal code areas, and the population was split into three sections based on the median annual net income of the inhabitants of the postal code areas. The median net income of the low-income areas was 13,948–17,131€/year, middle-income areas 19,212–21,303€/year, and high-income areas 21,359–26,907€/year. The residential areas were considered rural if the population density was < 76.9 inhabitants/km2 and urban if > 76.9 inhabitants/km2. The highest proportion of the population lived in high-income areas, and retirement rate and unemployment rate were higher in low- and middle-income areas (Table 1).

Statistical analyses were performed using IBM SPSS Statistics for Windows software. Proportional data are presented as numbers and percentages and were tested for significance using the Pearson’s chi-squared test. Continuous variables are presented as medians with 25th and 75th percentiles (25–75 PCT) and were tested using the non-parametric Kruskal–Wallis test. P < 0.05 was considered significant.

The crude rate of operation was calculated for each postal code area per 100,000 inhabitants per year. Age adjustments were performed by weighting each 5-year age group rate by the proportions of the age groups. A Cox regression analysis was performed to calculate multivariate hazard ratios (HRs) and 95% confidence intervals (CIs) for 5-year mortality. Variables with univariate significance < 0.1 were included in the model using the enter method. Variables with multivariate significance < 0.05 and those with a significant impact on the log-likelihood function were retained in the model.