Study design
Quantitative analysis of a retrospective cohort of patients who had bariatric surgery at the Hospital of Clinics from the University of Sao Paulo (HC-FMUSP), Brazil, from January to December of 2018, through Interrupted Time-Series Analysis (ITSA) on direct health care costs and health outcomes.
Bariatric Surgery Characteristics
The patients included in the study sample were distributed in three groups of surgery: Roux-en-Y gastric by-pass (R-YGB), vertical gastrectomy, and adjustable gastric banding. In most cases, open surgery was performed; however, a minor proportion of patients had surgery through video laparoscopy.
According to the standard protocols for bariatric surgery within SUS, there is a requirement for previous assessment of patients' eligibility for bariatric surgery in primary health care facilities.
Depending on the health status, patients are referred to a specialised health care unit.
Furthermore, patients are required to perform numerous exams and consultations pre-and post-surgery on a regular basis in Brazil.39
Patients with moderate to severe obesity diagnosis are referred to high complexity hospitals, like the HC-FMUSP, and included in the waiting list for bariatric surgery, performing monthly clinical and laboratory exams. After monitoring on eligibility criteria during variable periods, patients are submitted to surgery, hospitalisation, and post-surgery follow-up, starting a post-surgical period of monthly clinic and laboratory exams (Fig. 1).
Sample
Data on a cohort of 114 patients who had bariatric surgery at the HC-FMUSP in Sao Paulo, Brazil, through the Brazilian Unified Health System (SUS) from January to December of 2018, were obtained from the hospital's electronic medical records.
Only patients with complete data registered on the medical records were included in the study, encompassing information on multiple anthropometric, hemodynamic, and biochemical parameters through regular assessments of patients' health status, and utilisation of resources and costs of health care procedures, within six months pre-and post-intervention (bariatric surgery).61
Individual information registered on daily-based electronic data collection at HC-FMUSP were gathered in a single dataset encompassing data on patients' characteristics, health outcomes, outpatient health care (pre-and post-intervention), and inpatient health care, including detailed information on utilisation of resources throughout the screening, intervention, and follow-up.
Variables
The following patient information at baseline was obtained in medical records considering the periods of 180 days (6-months) pre-and post-intervention (Table 1).
Table 1
Information on health outcomes and utilisation of health care resources of patients from HC-FMUSP. Sao Paulo (Brazil), 2018.
Variable
|
Components
|
Health outcomes
|
Anthropometric measures (weight and height);
Hemodynamic measures (blood pressure);
Biochemical exams (cholesterol and fractions, triglycerides, insulin, glucose-linked haemoglobin, and fasting glucose).
|
Health care costs
|
Outpatient health care:
• Appointments with health professionals;
• Anthropometric measures (weight and height);
• Hemodynamic exams (blood pressure);
• Biochemical exams (cholesterol and fractions, triglycerides, insulin, glucose-linked haemoglobin, and fasting glucose).
Inpatient health care:
• Hospital length of stay;
• Type of surgery;
• Use of resources including the operating room, medication, meals, human resources, hemodynamic and biochemical exams.
|
The comparison of health outcomes and health care costs was based on the measurement of changes concerning the intervention's baseline (first day of hospitalisation for bariatric surgery).
Patients' demographic and lifestyle characteristics were also gathered to comprise control variables in statistical analysis, including gender and age, tobacco use, and alcohol consumption.
Health Outcomes
Information on health and nutritional status of patients, referring to the assessments pre-and post-surgery, were extracted in electronic medical records considering its associations with bariatric surgery in the literature5,23,25−31, comprising the following set of health outcomes adopted for statistical analysis:
Measurements of health outcomes during screening, intervention, and follow-up were performed by trained health professionals, according to standard procedures internationally recommended adopted within HC-FMUSP facilities.
Health Care Costs
Data on utilisation of resources during outpatient and inpatient health care were used to estimate patient's direct health care costs referring to bariatric surgery, and 6-month pre-and post-intervention periods, adopting the health system perspective through the micro-costing approach.
Prices of inputs and wages of health professionals involved in health care procedures were obtained from the HC-FMUSP institutional database, based on information on inputs purchases and human resources payroll.
Prices per item were multiplied by the amount used for the patient's treatment, and hourly wages were multiplied by the amount of time dedicated to performing procedures and consultations during the patient's treatment. Monetary values were updated to January 2020 and converted into US dollars using the Brazilian Central Bank official exchange rate.
Statistical analysis
Descriptive statistics and interrupted time series analysis (ITSA) with generalised estimating equations (GEE) and marginal effects were performed using single-centre retrospective data on costs and multiple health outcomes related to bariatric surgery. The information gathered for the sample of patients from the HC-FMUSP was split into two segments for analysis, i.e., health outcomes and health care costs before and after bariatric surgery, respectively.61
Dependent variables included in the models were health care costs and health outcomes (weight, BMI, blood pressure, cholesterol and fractions, triglycerides, insulin, glucose-linked haemoglobin, and fasting glucose).
GEE was fitted with uneven distribution for different outcomes during pre-intervention, intervention, and post-intervention to adjust monthly trends according to patients' characteristics. Marginal effects were obtained after GEE estimation by sample means at each period of evaluation (pre-and post-intervention) and used to estimate incremental health care costs and effects for each health outcome.
Interrupted time series ordered logistic models were estimated for health outcomes, and Poisson models were estimated for health care costs, controlling for age and gender with random effects estimator.
ITSA regression model uses a time series of a particular outcome of interest to establish an underlying trend interrupted by intervention at a given known point in time.
The model's statistical design draws an expected trend in the hypothetical scenario without the intervention, compared with the new trend established post-intervention to identify potential differences throughout time. The post-intervention scenario provides a comparison for the evaluation of the intervention impacts by calculating the change in slope throughout time, according to the following standard equation:62,63
Yt = β0 + β1 Δt + β2 Xt + β3Xt Δt + εt [Equation 1]
Where Yt is the accumulated result measured at each spaced time point t, Δt is the time since the start of the study, Xt is a dummy (indicator) variable representing the intervention (pre-intervention periods 0, or 1), and Xt Δt is an interaction term.63 The regression coefficient for Tt represents the rate of change of activity in stage 1, and the sum of regression coefficients for Tt and (XT)t is the rate of change of activity in stage 2. The effect over time was defined as the difference in the rate of change from stage 1 to stage 2, that is, the regression coefficient of (XT)t.
Thus, interrupted time series models were achieved by defining independent variables Δt = time point (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13), Xt =0 for time points in stage 1 (time points 1, 2, 3, 4, 5 and 6) and 1 for time points in stage 2 (time points 7, 8, 9, 10, 11, 12 and 13), and (XT)t =0 for time points in stage 1, and (t − 4) for time points in stage 2.
The intervention's immediate effect was defined as the regression coefficient corresponding to Xt (corresponding to the counterfactual difference between stage 1 and stage 2 evaluated at time point 7). The interrupted series time models for health costs and outcomes were ordinal logistic repeated measures models, including additive effects for Tt, Xt and (XT)t, and adjustment covariates. The statistical analysis was conducted using Stata version 14, and Newey-West standard errors were reported to account for autocorrelation at lag 1.59