Study design
We performed a cross-sectional secondary data analysis of a public dataset from the National Survey on User Satisfaction of Health Services 2015 (ENSUSALUD-2015). It was recollected by the National Superintendence of Health (SUSALUD, in Spanish) and the National Institute of Statistics and Informatics (INEI, in Spanish). ENSUSALUD-2015 is a national-wide survey carried out in Peruvian ACFs in 2015, which aims to assess the ACFs’ users’ perception about the care provided.
Context
ACFs are health facilities which administer health services to individuals who do not require hospitalization into a health care facility. In Peru, ACFs belong to any of four health systems: 1) Ministry of Health and Regional Governments (MOH-RG), which provide health care mainly to low-income people. It is supported by the Peruvian Government, and it works in three different levels: Local, regional and national; 2) Social Security (EsSalud), financed by the Ministry of Labor, which provides health care to formal workers, former workers, and their relatives. It operates using its own centers of health; 3) Armed Forces (Marines, army and aviation) and Police, financed by the Ministry of Defense and the Ministry of Interior, respectively. All of them works independently from each other. They provide health care to armed forces and police members, and their relatives; 4) The private sector, financed by users with private insurance or by paying at the same moment of attendance [
21].
ACFs are divided into three different levels: Level I ACFs provide basic health care and takes care of the most frequent low complexity problems. Level II ACFs have the capacity to resolve some surgical problems and the ones transferred from level I ACFs. Level III ACFs are reserved for the attendance of complex conditions requiring specialized medical procedures [
24]. In most cases, with the exception of the private sector, patients should be referred from lower levels in order to assist to higher levels ACFs. The staffing in the ACFs depends on the level of ACFs, but in all cases, the patient is attended by a medical doctor.
Variables
Outcome: patient satisfaction
The outcome for this study was patient satisfaction. It was evaluated using the question: “Regarding the health care service received today in this health facility, how would you rate your satisfaction level?” (“Respecto al servicio recibido el día de hoy en este establecimiento, ¿Cómo calificaría usted su nivel de satisfacción?”, in Spanish). It had the following response options: Very satisfied, satisfied, neither satisfied nor dissatisfied, dissatisfied, and very dissatisfied. All questions in the survey were asked in the patient’ maternal language.
For the purpose of the analysis and because of skewed data distributions, this item was dichotomized into two categories: “Satisfied” (if the participant answered “very satisfied” or “satisfied”) or “Not satisfied” (if the participant answered “neither satisfied nor dissatisfied”, “dissatisfied”, or “very unsatisfied”). Previous studies have used this methodology [
5,
26].
Exposures: waiting time and consultation time
The exposures of this study were waiting time and consultation time. Both were analyzed as continuous variables. To evaluate waiting time, we considered the following questions: “At what time did you arrive at the health facility?” And “At what time did you enter the physician’s office?” Both variables were recorded in minutes. The difference between these two times (time of entry to the physician’s office minus the time of arrival to the health facility) was defined as waiting time.
Consultation time was measured with the question: “How long was the time, from the moment you have admitted the physician’s office to the time you left the physician’s office?” This variable was recorded in minutes.
Other variables
Sociodemographic variables included in the analysis were: age (in years), sex (male or female), having finished secondary education (yes or no), wealth (measured with the question: “How much is, approximately, the monthly family income?” Later, this numeric variable was categorized in its 1st/2nd quintile [with the lower income, between 80 and 1500 PEN] and 3rd/4th/5th quintile [with the higher income, between 1501 and 50,000 PEN]) (1 PEN = 0.30 USD, approximately), having a chronic illness (evaluated with the yes/no question: “Do you suffer any illness for which -according to the physician- you require medical evaluations at least every three months?”), and having health insurance (yes or no).
Consultation variables included in the analysis were: Having a scheduled appointment (evaluated with the yes/no question: “Did you have a scheduled appointment for the health service you received today?”), being accompanied (evaluated with the question “Have you come accompanied or alone?”), and reporting that the physician explained his/her health problem (evaluated with the yes/no question “Did the physician clearly explain your illness or health problem?”).
Facility variables included in the analysis were: ACF level (I, II, o III), health system (MOH-RG, Social Security, Armed Forces and Police, or private), and geographical region (coast, highlands, jungle, or the city of Lima).
Data analysis
Data analysis was performed using STATA® version 14.0 (STATA Corporation, College Station, Texas, USA). We followed the ENSUSALUD-2015 sampling specifications, including stratification, expansion factor, and primary and secondary sampling units.
For descriptive analysis, absolute and relative frequencies were used for categorical variables, while mean with 95% confidence intervals (95% CI) were calculated for continuous variables. Mean waiting, consultation time, and patient satisfaction were calculated by each health care level, health system, and geographical region. Additionally, we used chi-2 test for bivariable analysis between patient satisfaction and health care level, health system, and geographical region.
Before undertaking the association analysis, we grouped the waiting time and consultation time variables, into 10 min intervals. Because we considered that the 10-min unit would be more helpful for the statistical and practical interpretation of the results, instead of the every-minute unit. In addition, since we considered an average waiting time of 90 min, 10 min intervals were considered appropriate. A similar approach, using minutes interval, were published elsewhere [
12]. Then, to evaluate the relationship between independent variables (waiting time and consultation time, both evaluated by 10-min intervals) with patient satisfaction, crude and adjusted logistic regressions models for complex survey sampling were fit to estimate odds ratios (OR and aOR) and their 95% confidence intervals (CI). Logistic regression models were adjusted by age, sex, having finished secondary education, wealth, having a chronic disease, having health insurance, having a scheduled appointment, being accompanied, reporting that the physician explained his/her health problem, health care level, geographical region, health system, waiting time, and consultation time. To avoid the influence of outliers, we excluded observations with consultation time > 30 min in the consultation time/patient satisfaction associations (19 participants excluded, 0.14% of total population). Their responded consultation time ranged between 34 and 90 min. This was because we considered that a consultation time with more than 30 min is a very rare scenario, and including these observations in the regression analysis would bias the results. We didn’t exclude very short consultations times because we considered those were plausible in our context. Additionally, very short consultations are not likely to affect the regressions coefficients since they are close to the mean consultation time, unlike very long consultation times.
In addition, the associations were plotted using restricted cubic splines with five knots obtained by default, as previously suggested [
27], in order to evaluate linearity and identify possibly cut-off points to difference subpopulations in which the association would be different.
Finally, we assessed the association between time and consultation time, using Pearson correlation test and, additionally we calculating the mean and 95% CI consultation time by each tercile of the waiting time.