Predictability and Financial Sufficiency of Health Insurance in Colombia: An Actuarial Analysis With a Bayesian Approach

Abstract

Every year, the Colombian government provides a prospective premium, known as the capitation payment unit (CPU), for each affiliated person (according to sex, region, and age) to each health insurance company, in order to manage the corresponding risk in health. This article studies the prediction capacity for the health expenditure for the more than 20 million affiliates to the contributory regime of health, as well as the CPU’s financial sufficiency, using an actuarial approach. Using the pure risk premium method and generalized linear models, both classic and Bayesian, the CPU is estimated; these results are compared to actual expenditure by an index of forecasting ability. It is concluded that the use of historical information about expenditure on health, as well as the Bayesian inference, among the other methodological innovations developed, provides an advantage for obtaining more accurate prospective values. These technical recommendations seek to support an improvement in the public budget allocation of more than 6 billion dollars per year to the Colombian health system.

ACKNOWLEDGMENTS

We express our gratitude to Oscar Melo, Mauricio Martínez, Paul Rodríguez, Juan-Camilo Vargas, Laura Mora, Sebastián Higuera, Piedad Urdinola, Raúl Macchiavelli, Jhonathan Rodríguez, Adriana Robayo, Jaime Ramírez, Giancarlo Buitrago, Mery Bolívar, Freddy Hernández, Alejandra Sánchez, Diego Ávila, Esteban Orozco, Giancarlo Romano, Angélica Ordoñez, Armando Zarruk, Kelly Estrada, and Oscar López, for their useful comments and suggestions.

Notes

1 A health technology “is an intervention developed to prevent, diagnose or treat medical conditions; promote health; provide rehabilitation … The intervention can be a test, device, medicine, vaccine, procedure” (O’Rourke, Oortwijn, and Schuller 2020).

2 Although national health systems are different, unique, and sui generis between countries, it can be said that structural pluralism was adopted as the basis in Colombia (Londoño and Frenk 1997). However, the capitation mechanism explained here is unique at the international level.

3 This study does not evaluate the subsidized regime, given that public information about its health expenditure is inadequate because of the poor quality of the records of the health insurance companies that belong to this regime.

4 The advantage includes the ability (i) to incorporate available data in estimating the initially given parameters, (ii) to allow the distributions of the analysis parameters to be specified, and (iii) to minimize the use of arbitrary limits to make decisions, among others (van de Schoot et al. 2021).

5 To obtain the number of people exposed to the risk in $t+1,$ they use the projected growth rates of the population from the National Administrative Department of Statistics (DANE, its acronym in Spanish).

6 The GD only has the reports of health technologies from health insurance companies whose data satisfy the validation meshes of the MHSP. For our analysis period, on average, this includes more than 80% of the affiliates of the CR, which means over 18 million people. The details contained in these data are, at the municipal level, simple age, EAPB, health technology, and International Classification of Diseases 10th Revision (ICD-10).

7 The average exchange rate of COP to 1 USD for each year is: 2013, 1869 COP; 2014, 2000 COP; 2015, 2744 COP; 2016, 3051 COP; 2017, 2951 COP; 2018, 2956 COP; 2019, 3281 COP; and 2020, 3693 COP.

8 These are health technologies that were not initially financed by the CPU but at least have a homologue that is financed. In this context, the health insurance companies can recover the difference between the values of these two health technologies.

9 Since the information is annual, the most recent data available in year $t$ are those referring to $t-1.$ That is why there is a lag of one year in the actuarial calculation, from $t-1$ to $t+1.$

10 For approximation (c), the Bayesian GLM, the median of the a posteriori distribution of each parameter is chosen.

11 The dummy variables of Period are set to zero so that the fixed effect for any year of reference is not considered in the estimation.

12 The empirical premium is understood to be the actual expenditure on health (in finance technologies for the CPU) that occurred for a certain age group in a certain type of region for a specific year.

13 These values are extracted from the resolutions of the regulatory entity (Ministerio de Salud y Protección Social 2015b; 2016b; 2017b; 2018b; 2019b). Given that the value presented there is of the commercial risk premium, this is multiplied by 0.9 to obtain the pure risk premium (by law, the administrative margin that the health insurance companies of the CR hold is 10%) (Ministerio de Salud y Protección Social 2011).