Algorithms as decision-making aids

Pilot mit Checkliste als Entscheidungshilfe
© David (CC BY-NC-SA 2.0), via flickr: Pre-flight Checklist

Simply stated, algorithms are procedures that govern processes. They can be found online in the form of product recommendations, in digital camera chips, or in calculating the fastest route or the cheapest flight for a trip. Sometimes these algorithms appear complicated and intransparent or even questionable when we feel manipulated by them. However, there are also very simple algorithms that can provide guidance in complex situations.

One such algorithm is encountered, for instance, in first aid courses. Participants learn a series of interventions to be used in an emergency. Note that this is an algorithm that is applied by people instead of computers. In the event of heart failure, modern defibrillators also have an algorithm that gives the person performing first aid exact instructions on how and at what speed resuscitation should take place.

A wide variety of other possible uses for algorithms in medicine exist. They can for instance be relevant when patients go to the emergency room with unclear symptoms: On the basis of these few symptoms, the physician then needs to quickly make a decision whether to send patients home, treat them immediately, or refer them to other specialists. Here one needs a structured procedure that is used consistently in order to eliminate sources of error and achieve reliable results.

As research has shown, just as pilots review a checklist before starting their planes in order to increase air traffic security, so can automatically using behavioral guidelines increase safety in medicine and lead to better treatment. Doctors, who like all people can become tired or distracted in the course of a workday, are thereby offered support that makes it easier for them to reach a diagnosis, which benefits the patient.

Unfortunately, however, many are skeptical about using algorithms in the medical domain. Primarily patients feel insecure or less confident in physicians' competence when the latter resort to decision aids. They are concerned that the algorithm in a computer program can make only general recommendations that do not apply to their individual case. That is where the misunderstanding lies: precisely by collecting all relevant facts on a concrete patient (e.g., sex, age, symptoms, blood test results, genetic information, previous illnesses, medication) is an algorithm capable of making a very individualized statement. Those who make use of such computerized decision aids are in fact better, not worse doctors.

The Harding Center investigates how to develop transparent and practicable algorithms for everyday clinical use that both doctors and patients are comfortable with and that ideally combine medical experience and intuition.

We represent such algorithms in simple decision trees (also known as fast-and-frugal trees).