Securra & Cognitive AI

Securra’s number one goal is improving the healthcare outcomes over patients everywhere. Better health outcomes arise from better care. But in most cases, care is applied reactively. Providers can intervene only after a patient presents with serious symptoms. And even in these cases, they may not have all information required for good decision making available to them.

Securra seeks to reverse this pattern and move healthcare toward a proactive model of care. To do this, we require a platform that allows for integrated and up-to-the-minute information on a patient’s health status. We need to collect and analyze data continuously with the latest techniques from machine learning to spot emergent conditions and anomalies. And we must apply cognitive reasoning to provide targeted recommendations for action, with transparent and explainable justifications.

This model allows Securra to measure, understand, decide, and act proactively in order to foster improved provider reach and quality of care. In a world with growing health care needs, this mission is more urgent that ever.


Applying analytics in healthcare can be difficult because required static data is often siloed. This means that information on a patient is spread across different institutions and file types. Platform-based solutions like Securra provide a location for integrating broad, heterogenous data on a patient in one place. This breadth of data, including demographic info, patient histories, and social determinants of health, provides necessary background for informing analysis of continuous patient data.

In addition to static data described above, to provide actionable diagnostic or predictive analytics typically requires access to more continuous data. On a typical schedule, your doctor might check your blood pressure once every six months. But a much more accurate picture of the dynamics of your health can be established by measuring this continuously. Numerous wearables and smart devices can be used to collect this information in real time. It is then merged with other historical information. While many RPM solutions are “closed” Securra allows many devices to work with its platform.


The combination of demographic and continuously collected patient data can be used to develop machine learning systems that produce valuable insights. Insight can be as simple as the detection of an anomaly (like a patient’s blood pressure varying in an unobserved way throughout the day) or more complex like identifying a particular condition (such as cardiac arrhythmia). In addition to building detection systems, techniques can be applied that provide predictive analytics. In these cases, the emergence of a particular event could be predicted to occur in the next several hours if there is no change in state.

While machine learning techniques can provide powerful applications, they are not without their pitfalls. One issue is that there must be sufficient data to train algorithms. This means not just sufficient quantity but also high-quality data. Not all data is created equal. We need to ensure we have consistently collected the data and have good coverage of the ranges of data we expect to encounter in “real world” settings.

Even when these factors can be controlled, there are still limitations we must be aware of when it comes to machine learning approaches. An important thing to keep in mind is that these algorithms learn in a data-driven fashion. They essentially learn to match patterns that emerge from nuanced statistical attributes of the data, but may not understand how these patterns relate to other phenomena.

So, while it is fair to ask a machine learning algorithm “what is happening” it can be more difficult to ask it “why is this happening.” In practice, this means providers can use the outputs of machine learning systems but must do the work themselves to contextualize these insights when searching for appropriate interventions. This relates to another problem that most machine learning approaches are “black boxes.” We cannot often say what particular evidence or factors are driving the algorithm’s recommendations. This can be especially problematic in high-assurance areas such as healthcare.


While machine learning certainly is powerful its foundations being based on large datasets and statistical induction invites several pitfalls. However, there is another approach to artificial intelligence that can help overcome these challenges. While machine learning can be said to be statistics driven, the complement to this approach is logic driven AI.

Early in AI’s history, logic-based AI was the dominant approach. The techniques were so successful and many original solutions are still in use today, to the point that we no longer call these original approaches “AI.” The benefit of these systems is that rather than learning from data, they can be programmed explicitly with the knowledge they need to know. This knowledge can be used with logical analysis (reasoning) on new data to make new inferences. This is the basic of the expert system and is potentially very powerful in healthcare where it can (1) make use of explicit knowledge like physics concepts and care pathways (2) provide clear explication of any decisions by exposing the evidence and reasoning used to produce them.

The original logic-based systems did have some drawbacks. One problem is that manually programming knowledge can take considerable time. And of course, this is impossible for a number of problems. Imagine programming the explicit rules that let you identify a person’s face; compare this with simply feeding data to a machine learning system. Another short coming is that early logic-based systems were very brittle and could only reason deductively. This means they were unable to create new knowledge using methods like induction and hypothesis and so they could not learn from feedback from the environment.

Today, these logic-based systems have come further and are able to automatically update themselves when new facts come to life; this makes them much more powerful and an excellent match for use in healthcare environments where providers can examine the logic of the cognitive system and provide positive or negative feedback to any new discoveries the system may make (confirm or deny the new knowledge before letting it be used in practice). But the most powerful types of system are those that combine machine learning with cognitive reasoning.

This hybrid approach to AI is what sets Securra apart from other healthcare companies. At Securra, machine learning systems are used to collect valuable “clues” about the state of a person’s health. Cognitive reasoning is then deployed to pieces these clues together into a holistic picture and reason from first-principles about the best way to intervene. The result is that Securra systems provides doctors with an actionable recommendation with a transparent audit trail that lets them validate the reasoning before commit to a particular intervention. The result is a robust, flexible system that improves the reach of providers and positive outcomes for patients.