Data analytics can be defined as the science of examining a dataset with the aim of drawing conclusions from it in order to assist decision-making, or simply to broaden the understanding of particular issues.

Accordingly, data analysis involves a range of techniques used to derive conclusions and recommendations from data. The various data analysis techniques fall into four main categories, based on the insights they provide: descriptive, diagnostic, predictive and prescriptive.

Descriptive analytics: This type of analysis is usually conducted using a large dataset which, on the face of it, is not very informative. Cleansing, ordering, processing and visualisation techniques are applied to illustrate developments within organizations in an accessible format. Descriptive analytics produces an analysis of the outcomes of particular events or phenomena and can identify what happened. The outputs simply indicate whether or not something is going well, without explaining why.

Diagnostic analytics: This provides detailed information about a specific issue and identifies the reason behind a particular event or phenomenon. Historic data can be compared to other data to answer the question as to why something happened.

Detailed information is required to conduct this type of analysis. Otherwise, the data collection required for each specific issue may turn out to be inefficient, as well as time-consuming.

Predictive analytics: This type of analysis applies mathematical and statistical modelling and techniques to historical data held by the organization. Predictive analysis cannot determine with absolute certainty what will happen in the future, since the analysis is based on levels of probability, but it does show what might be expected to happen. It draws on the outputs of descriptive and diagnostic analytics to identify groups and outliers, which are used as a basis for the predictive models. These models, applied to a large quantity of data, can produce a forecast of what might happen to a given degree of probability.

Predictive analytics is an advanced form of data analytics which brings with it many advantages, such as the use of machine learning. However, it is important not to lose sight of the fact that a forecast is only an estimate. Since it is based on an analysis of the correlation between variables and how they might develop, its accuracy depends to a large extent on the quality of the underlying data.

Prescriptive analytics: This type of analysis involves collecting data, recommending actions and predicting the impact they will have, in order to facilitate and automate decision-making by identifying the best decision from a range of options.

It aims to respond to questions such as “What can I do to stop this from happening?” or “What can I do to make this happen?”. It sets out measures to take to prevent a future problem or to make the most of a trend.

This kind of analysis requires not only internal historical data, but also external information, because of the nature of the underlying mathematical algorithms.

Figure 1: Different types of data analytics

Figure 1 summarizes the categories, their added value and their complexity in terms of implementation.

The use of these techniques is especially important in the field of social security, since they are enabling institutions to make the most of the increasingly large volume of data available, both to detect and explain unusual events, and to build predictive models which can help to anticipate new developments.

This type of analysis has many applications. A growing number of institutions are using analytic technologies in a range of fields, such as fraud prevention, process performance analysis, the evaluation and adjustment of social programmes, the implementation of preventive measures, the proactive development of social policy, and healthcare services.

The International Social Security Association (ISSA) has developed guidelines on information and communication technology, which include guidance on the application of data analytics technologies. Specifically, there are four guidelines, one for each category of data analytics set out above (guidelines 54 to 57), as well as separate guidelines on big data (guideline 58) and machine learning and its use in supporting decision-making (guideline 59).

Experiences of applying analytical techniques to combat the pandemic

Examples of good practice in this area have been presented during the competition for the ISSA Good Practice Awards for the Americas 2020, and in webinars and other ISSA activities. It is clear from these examples that analytical techniques are an area of current interest and are increasingly being applied by social security institutions. During the COVID-19 health emergency these techniques have proven to a be valuable tool, with particular applications in healthcare services.

Costa Rica

The Social Insurance Fund of Costa Rica (Caja Costarricense de Seguro Social – CCSS) applied analytical techniques to track the spread of COVID-19 in the population and the services provided in CCSS health establishments. During the health emergency, there was a clear demand for timely and accurate information both for tracking purposes and to support decision-making to manage the pandemic.

For this reason it was decided that a single information source should be developed, drawing on the robust and adaptable health information available to CCSS in order to conduct a traceability analysis of COVID-19 patients. To this end, it was necessary to consolidate data from health establishments and draw on the institution’s own master data.

The resulting application also places a strategic focus on the use of business intelligence tools to manage statistical information, enabling the timely provision of COVID-19 healthcare data. The aim is to foster data-based decision-making, in particular at a strategic level. This requires easier access to statistical information as well as the use of data analysis tools.

Overall, the application of analytical technologies did help CCSS to make decisions based on the data. CCSS was able to quantify the impact on services and adapt provision accordingly. At the same time, the application supported the institution’s own digital transition, in particular through the adoption of data analytics technologies.

The CCSS approach is based on descriptive and diagnostic analytics.

Mexico

In Mexico, the Mexican Social Security Institute (Instituto Mexicano de Seguridad Social – IMSS) has addressed the issue of data analytics by adopting a data governance strategy, in view of the importance of data in the COVID-19 pandemic.

The data governance approach means that IMSS can provide valuable information to its key constituents. This facilitates decision-making at both the management and the operational level. The geographical and statistical information available to the Institute enables it to manage levels of stocks, beneficiaries’ incapacities caused by ailments related to COVID-19 and company-related data regarding compliance with return to work variables. Through this approach, IMSS is seeking to address issues such as duplication of effort, missed opportunities to deliver valuable information, and the poor standardization, traceability, integrity and quality of data. These features are typical of data from institutions which provide multiple services and hold large volumes of information across a number of unconnected data repositories.

The IMSS’s data governance model is based on three pillars: a cultural shift on the part of individuals, improved processes due to a more holistic vision, and technological development, with the implementation of a data lake as part of an integrated solution. With this foundation in place, analytical tools can then be deployed to carry out data analysis. Data analytics provides useful, high quality information and enables the identification of aberrant trends and values. It facilitates the harmonization required for a transparent and secure exchange of information with other institutions.

Again, in the same context, the harnessing of big data, based on a data lake, makes it possible to visualize and manage the evolution of the pandemic, applying variables such as the number of cases, hospital occupation, patient outcomes and deaths per region, sex, age, medical unit, etc.

The IMSS approach is based on descriptive and diagnostic analytics.

Peru

The approach of the Peruvian Social Health Insurance Institute (EsSalud) was to set up a business intelligence and data analytics unit, which aims to provide complete and timely, high-quality information, using data analysis to underpin institutional strategic decisions. The creation of this unit is intended to encourage modern, efficient management in the interests of beneficiaries, by providing senior management with timely, relevant and high-quality information to support decision-making.

Another aspect of this approach is to develop applications and other innovative strategies to communicate with senior management, as well as to promote and monitor better quality record keeping and the integration of institutional information. One of the key applications is a heat map (Mapa de calor), which shows the development of the pandemic across Peru, plotting the “route of the virus” as it spreads, high prevalence areas, etc.

The supply of relevant information, updated daily, makes it possible to analyse postponed appointments, track home visits, send alerts about extended admissions and monitor the availability of hospital beds.

It is worth emphasizing that EsSalud aims not only to generate information for internal use, but also to cooperate with other public bodies, sharing the outputs obtained. Information has been used, in particular, from the National Registry of Identification (Registro Nacional de Identificación) and Civil Status (Estado Civil).

The approach of EsSalud is based on descriptive and diagnostic analytics.

Results obtained

These institutions have achieved positive results thanks to the application of analytical technologies. They have been able to implement specific mechanisms to address the challenges of the pandemic and have also increased their ability to apply these technologies in other contexts. Table 1 summarizes the results.

Table 1. Results obtained in various countries

Country	Results obtained for healthcare systems
Costa Rica - CCSS	Strengthening decision-making processes.
	Generation of information to help plan post-COVID-19 services. It was possible to measure the impact on services, which was: 26% for external consultations 35% for emergency wards 34% for hospitalization 55% for surgery
Mexico - IMSS	Implementation of a unique data platform for COVID-19 using data lake and big data technologies. The platform has received 86,422 visits by 445 users.
	Information is shared with the Ministry of Health (in charge of pandemic policy coordination) and other institutions.
Peru - EsSalud	Daily pandemic monitoring reports on the pandemic were provided, tracking: Positive COVID-19 cases (geographical location, rate per 1,000 inhabitants); Positivity rate per healthcare network; Updated hospital bed status; Update on supplies of strategic goods.
	Implementation of applications such as the heat map (Mapa de calor), as well as maps and infographs to assist with the interpretation of information. Setting up a dashboard providing alerts on extended stays and hospital bed availability, available online with information updated in real time.
	Information is shared with at least 12 local governments and two ministries.

The potential and limitations of predictive analytics

The Pan American Health Organization (PAHO) emphasizes the importance of predictive analytics in the fight against COVID-19, since it allows us to estimate the pandemic’s behaviour within an acceptable degree of uncertainty. On the basis of this information, institutions can anticipate the approximate demand for acute medical services, determine the timeframes for partially or fully lifting containment measures (i.e. lockdowns), and even predict new needs that may arise (PAHO, 2021).

At the same time, PAHO also points out that forecasting models have some limitations when applied to a context such as the pandemic, since there is some inherent uncertainty in the models which affects their performance and can be difficult to quantify. In particular, the introduction of time horizons and the heterogeneity of the data being analysed can lead to greater uncertainty. Performing a “sensitivity analysis” is key to better understanding uncertainty. This technique is used to evaluate the impact that a particular dependent variable, under a given set of assumptions, might have on the overall result. Uncertainty can be reduced by increasing the sample size and improving the quality of data used in the model. This means that the volume and quality of data is critical to these analytical techniques (PAHO, 2021).

Critical factors

The experiences described all point to various critical factors when applying analytical techniques in the context of COVID-19.

A first critical factor is the team, which is a vital part of the strategy to implement analytical techniques. For this reason, the aim should be to create multidisciplinary teams with clearly defined roles. It is always important that committees should include members from different areas since this makes it possible to extend the definitions when necessary.

In EsSalud, a specific analysis unit has been set up, comprising a multidisciplinary team. This unit has a flexible strategic plan, allowing it to develop and add new innovations to the institution’s analysis strategy. The IMSS in Mexico put together a joint team for collaborative work. Specific roles were defined in the areas of analytics, data quality, architecture and the functional ownership of data. An executive committee for data was created.

Support from senior management is also crucial to this type of initiative, in particular to ensure viability in implementation. This is because data analytics initiatives usually cut across institutions, bringing together data and processes from more than one area of interest. They may also require agreements to be reached with other organizations. Both IMSS in Mexico and CCSS in Costa Rica were able to rely on the engagement and support of senior management, which was key to the success of both initiatives.

The flexibility of the models implemented can also be regarded as a critical factor. These models must evolve to reflect changing realities. It is also important that the models should be specific, not general, since this improves the forecasting performance. In the case of EsSalud, the versatility of the heat map model meant that it could be adapted to the growing need for information arising from the health emergency.

Another critical factor is the quality of data, which is the main basis of any data analytics initiative. Data-based decision-making requires reliable data, otherwise the power of the analysis and the validity of the conclusions is limited. The best way to guarantee the quality of data is to take steps to ensure that good quality data is input from the outset, avoiding the need for data cleansing at later stages.

Both the CCSS in Costa Rica and EsSalud in Peru developed a culture of data-based decision-making. Action was taken to compensate for any lack of experience or familiarity with this approach, since it was understood that the quality of data and the correct interpretation of information are vital to ensuring the right decisions are taken.

Conclusion

Traditional data analysis techniques allow for automatic reporting and the creation of dashboards that can provide a retrospective view of the organization, in order to answer questions such as “what happened” and “why” did this situation arise. However, in addition to assisting decision-making by providing a descriptive analysis of data, advanced machine learning techniques such as predictive and prescriptive analysis can provide a future-oriented perspective of an organization, supporting decision-making, while at the same time optimizing business processes and increasing productivity.

The importance of data analysis has increased over recent years, as is apparent from the number of good practices that have been described, as well as the various webinar presentations on the same topic. To this end, data analysis tools have played an increasingly central role in organizations. Data analysis should not be considered in isolation, but rather in the context of business processes and decision-making, as well as the management and quality of the underlying data to ensure that the interpretation of data is well-founded.

It is also worth highlighting the institutional capacity of those organizations that already had data analysis projects underway, enabling them to re-focus their efforts in the context of the health emergency.

In summary, during the COVID-19 health emergency, a number of social security institutions have taken the opportunity to consolidate or start developing solutions of this kind.

References