Setting and participants
From June 2014 to December 2017, a cross-sectional study was conducted at the Hospital de Especialidades Eugenio Espejo (HEEE), a regional reference public hospital for endocrine neoplasia in adults in Quito, Ecuador. Ecuador is geographically divided into four major natural regions (Coast, Highland, Amazon, and Galapagos Islands). Due to HEEE being located within the Highland region, its patients come mostly from this area. All the patients who were seen for thyroid cancer at HEEE were included, except the patients who did not have the histopathology report. Patients who had initial management (including surgery) outside HEEE were also included.
Data collection and variables
Two sources of data were used to collect the variables of interest. First, a study coordinator interviewed eligible patients during their first postsurgical appointment at the endocrine clinic. During this process, the study coordinator captured: 1) demographic characteristics such as age, degree of education, region of residence (Coast, Highland, Amazon, or Galapagos Islands), age at diagnosis, and ethnicity; 2) family history of TC; 3) environmental risk factors; 4) methods of diagnosis (incidental or non-incidental findings). Second, study team members reviewed medical records of included patients to extract the following information: 1) thyroid gland functionality (euthyroid, hypothyroidism, or hyperthyroidism), thyroid ultrasound characteristics, and thyroid nodule fine-needle aspiration(FNA) cytologic results based on Bethesda System; 2) surgical characteristics such as type and extension of surgery; 3) thyroid gland histopathological features including tumor size, type, focality, minor or gross local invasion, and cervical lymph node involvement or distant metastases; 4) TC markers measured after thyroidectomy and before radioactive iodine therapy, including thyroid-stimulating hormone (TSH), stimulated thyroglobulin (sTg), inhibited thyroglobulin (iTg), and anti-thyroglobulin antibodies (aTg); 5) surgical characteristics such as type and extension of surgery, and complications (hypocalcemia <6 months and >6months after procedure, recurrent laryngeal nerve injury); and finally 6) the radioactive iodine treatment, its doses, and scan results.
Data management
Baseline characteristics data were managed as follows: employment and education were classified according to the National Institute of Statistics and Census (INEC) from Ecuador[11], and thyroid surgery settings were grouped as tertiary (hospitals providing specialized TC management) and non-tertiary hospitals. Furthermore, patients were considered to have a family history of TC when first and second generation-degree relatives had the disease. Based on thyroid histopathologic features, patients were diagnosed as medullary or non-medullary TC, the latter being further classified as differentiated (papillary and follicular), poorly differentiated, undifferentiated (anaplastic), or squamous cell carcinoma[12]. The risk of recurrence in differentiated TC was calculated by using the American Thyroid Association (ATA) 2009 risk stratification system, which classifies patients’ risk of recurrence as low, intermediate, or high[13]. Due to the overwhelming increasing incidence of patients with papillary thyroid cancer (PTC) with an intrathyroidal tumor size of < 1 cm, a new category was included to the ATA risk of recurrence calculator: “very low risk”[14]. Furthermore, the risk of mortality in patients with PTC was estimated based on MACIS score (metastasis, age, completeness, invasion, and size)[15]. A cutoff of 6 was employed to group patients as either low (MACIS < 6) or high risk (MACIS ≥ 6) of mortality.
Thyroid cancer method of detection was divided in two groups: non-incidental diagnosis (when the TC was found in a symptomatic patient) and incidental diagnosis when a thyroid nodule harboring TC is found during the workup of non-nodular thyroid disease, or during an imaging test requested for reasons unrelated to a thyroid disorder or symptom (e.g., preventive ultrasound), or TC is found incidentally in the histological examination of the thyroid gland removed for a benign condition (Figure 1) [16].
We classified the setting of the surgery as either tertiary hospital (HEEE and Hospital SOLCA) or non-tertiary hospital. Moreover, we evaluated the quality of thyroidectomy based on post-operative sTg levels (at least 6 weeks after the procedure)[17–19], and the frequency of surgical complications[20–22]. We considered that the quality of surgery was optimal when there were no post-surgical complications and when patients had a sTg ≤2 ng/dl, and poor when patients had at least one permanent surgical complication or post-operative sTg > 2ng/dL. Given that surgical complications and post-operative sTg levels could be affected by the presence of metastatic disease, we limited the assessment of the quality of surgical outcomes to patients with non-metastatic differentiated TC undergoing initial thyroid surgery (total thyroidectomy and prophylactic central neck dissection). Before 2016, the criteria for using iodotherapy included the ATA 2009 guidelines; after 2016, the ATA 2015 guidelines were considered.
Statistical methods
For categorical variables, frequencies and percentages were reported. For numerical variables, we used mean and median with their corresponding standard deviation (SD) or interquartile ranges (IQR), as measurements of central tendency and dispersion. Normal distribution was determined by visual inspection and by using the Kolmogorov-Smirnov test. Our dependent variables used for exploratory analysis were incidental findings and quality of surgery, which are dichotomous variables. For our bivariate and multivariate analysis, we decided to use prevalence ratio (PR) instead of odds ratios (OR) because PR is easier to interpret and OR tend to overestimate the results[23]. To calculate this PR, we planned to use a generalized linear model (GLM) with the binomial family and the log link. However, convergence problems were found with some of the variables. Such issues are common[24, 25]. At the end we chose, from all possible solutions, to use Poisson as the family for the GLM with robust variance. For the multivariate analyses, we decided to include in the models for incidental findings and poor quality of surgery all variables in which p-value was less than 0.05 and those considered to be important by the investigators. The results are reported as PR and their respective 95% confidence intervals. Statistical analysis was performed with STATA[26].