Open source software as a model for health care

by Andrew Oram

This was originally published on O’Reilly Media’s Strata blog, October 11, 2012.

(The following article sprang from a collaboration between Andy Oram and Brigitte Piniewski to cover open source concepts in an upcoming book on health care. This book, titled "Wireless Health: Remaking of Medicine by Pervasive Technologies," is edited by Professor Mehran Mehregany of Case Western Reserve University. and has an expected release date of February 2013. It is designed to provide the reader with the fundamental and practical knowledge necessary for an overall grasp of the field of wireless health. The approach is an integrated, multidisciplinary treatment of the subject by a team of leading topic experts. The selection here is part of a larger chapter by Brigitte Piniewski about personalized medicine and public health.)

Medical research and open source software have much to learn from each other. As software transforms the practice and delivery of medicine, the communities and development methods that have grown up around software—particularly free and open source software—also provide models that doctors and researchers can apply to their own work. Some of the principles that software communities can offer for spreading health throughout the population include these:

Some background on open source

The term "crowdsourcing" was introduced by Jeff Howe in WIRED magazine in 2006 and later became the title of a popular book by Howe. The idea of information sharing goes back to ancient times, through maxims such as "Who is wise? He who listens to everyone." But the power of the Internet and collaboration among volunteers to vastly increase the value of information was really revealed by the open source or free software movement.

Many computer programmers have shared their code freely from earliest times, passing it around on tape or whatever other low-cost, portable media were popular. Part of their motivation was simply that selling code was logistically more trouble than the proceeds would be worth: it would require setting up a company, creating a complicated licensing scheme, policing usage, and so forth.

But more fundamentally, programmers realized that sharing the code was of benefit to them. Those who took the code would submit bug fixes, improve the code, and add new features. To this day, most functioning computer systems are mash-ups of different cooperating programs from different developers, often open source. Proponents of the open source movement include Yochai Benkler, whose book Wealth of Networks is probably the best-known research on the topic, Eric Raymond, who in The Cathedral and the Bazaar wrote many popular aphorisms of the movement, such as "given enough eyeballs, all bugs are shallow," and Eric von Hippel, whose research at MIT showed that even in commercial industry, companies have taken many of their innovations from customers.

Open data is now a rallying cry for advocates of public information and more effective governments, but the lesson of the open source movement is that the mathematics and algorithms used to process data should be open as well. This applies to health care because data is inert in itself. Some kind of processing must be applied to extract useful information, and it this takes the form of open source code, many people can check it for accuracy, reuse it, and upgrade it. It is this continuous activity of sharing and upgrading that drives and defines quality and value over time. In other words, the community is the purveyor of value.

To start thinking this way in health, we can look at a simple example of a knee replacement operation. The individual undergoes the procedure and then relatively blindly enters the post-op recovery period. Is he experiencing optimal or sub-optimal recovery? What is reachable in terms of recovery given his age, gender, pre-op conditions and so on? Answering the question regarding the quality of his post op event stream will depend upon many folks sharing their experience in a timely manner such that patches or improvements can be uncovered while there is still enough time to implement them.

A good example of open sharing to nudge health event streams in optimal directions is PatientsLikeMe. However, this sharing behavior need not be limited to folks with chronic diseases. In health, we have no way currently of knowing if our post-op course or our health journey is "on track." Are we reaching what is reasonably "reachable"? Many conventions that support self-organization were developed in the open source and free software movement. Although a key principle is that anyone can offer software code updates (no contribution is too small), members of a team accept the need for a central authority or group of experts to keep malicious or poorly designed changes out. The term "benevolent dictator" often applies to the person or people whose authority to approve code is broadly accepted. Linus Torvalds, who invented Linux, now maintains it in consultation with many advisors. Again, as many elements change rapidly, the work to be done must spread to a wider base of experts.

Open health research

Innovations in health care are often controversial at first, and an open source attitude toward experimentation seems inconceivably reckless, especially in cases where the therapeutic window is narrow. In these cases, the cost of errors is high.

However, lifestyle management recommendations often enjoy a wide therapeutic window, so that adjustments in one direction or another are less likely to be dangerous. Evidence-based lifestyle management, learned through an open source attitude toward experimentation, may be an ideal way to teach free living communities the skills they need to protect against evolving threats. In this way the self-organized group becomes the trusted entity.

Perhaps the media may suggest that a cold breakfast cereal has the power to reduce cholesterol. Using an open source attitude armed with wireless health devices enabling home testing of cholesterol levels, the crowd can test this claim themselves. Social networking enables a few hundred friends to unite the quest of testing the" lowers cholesterol" claim. After 6-8 weeks of cereal ingestion and sharing pre and post cholesterol levels, this group can determine whether the claim is valid when applied to their free-living community.

This proposal is not entirely hypothetical. MBA students at the University of Mississippi launched work along those lines in the summer of 2010. Seventy-five MBA students with diverse backgrounds split into eight teams to tackle the obesity challenge in Mississippi. What was remarkable about their work was not only how different the eight proposals were, but how much effort and enthusiasm went into their work. Many were disappointed to have simply developed their projects on paper, in the form of power point slides and white papers. Unfortunately, the technology industry in 2010 had not yet matured to the point where students could easily and affordably mash up a solution and actually deploy it to their community.

Most technologists and scientists are familiar with how the National Oceanic Atmospheric Administration (NOAA) disseminates a valuable substrate of rich weather data, enabling a wide range of industries: weather reporting services, TV channels, economic forecasters and others, to generate their own business models.

The NOAA’s task is simple: drop a few sensors into the environment, collect the rich flow of synchronous data about many different weather and climatic parameters, stream the data into a data commons, and thus provide the fuel to power predictive modeling engines and to support a variety of different business models. The NOAA shares its sensor data freely with many users, who comb through the high-yield data and compare the weather outcomes of previous parameter co-occurrences, to better predict significant weather events and patterns. In this way, the ability to predict high-risk outcomes such as tornadoes, droughts and mass flooding is improved by all previous events. Old weather provides the insight to predict new weather.

In health, the task of prediction is more challenging. High-yield sensor data (activity, sleep, nutrition) are not currently being seamlessly streamed into a community data commons. However, in the era of wireless health, students can provide the much-needed resource to both start the data flow, then ensure the flow is sustained. Students everywhere are a uniquely scalable resource, able to hand-hold the older generation while they gather and transmit low-risk, high-yield health data.

The drive for students to participate goes beyond tuition fee reduction (a later reward perhaps) to more immediate rewards such as gaining a grade advantage or improving their resumes.

Ensuring optimal performance from highly-dynamic, complex systems is not a simple task. Assuming that a group of trained experts will keep up the skill set to ensure optimal performance over time has become unrealistic. Too many elements are changing too drastically, too quickly. Traditional safeguards for business models based on intellectual property rights may impede the evolution of new and better-functioning solutions. With our current pace of change, we will need to find ways to spread the work to be done beyond narrow expert groups to the wider skill set within the non-expert general community.

Innovative business models

Although open source goods are usually distributed over the Internet for no cost, they can be the center of a rich commercial environment. Linux, for instance, lies at the center of many company offerings, and Red Hat, whose offerings are entirely based on Linux and a lot of free software built to work with it, earned nearly three hundred million dollars in one quarter of 2012.

Open source data and software are also used to develop non-free offerings. Likewise, the community data commons modeled after the NOAA example can be the basis for a number of innovative business models.

The NOAA framework used by the University of Mississippi in the scenario described earlier has also been introduced to a number of campuses in the US, UK, and Europe. The hope is that each university will adopt a minimal technical framework for the data commons, such that the raw data (about sleep, physical activity, diets, medication effects, and more) moves relatively freely across campuses. Each campus is then able to build its own intellectual property on that data substrate.

Early examples might include advancing tools to mine ultra large data sets or advancing privacy-preserving layers for networked data, which is beyond the needs for un-networked data. In this way, each campus is expected to draw on the commons to develop their own unique areas of expertise.

There are at least four key ways to enable innovative business models:

In this way, just as in the NOAA example, value in the form of reliable data volume is created first; then, after a critical mass of data has accumulated, various business cases are enabled. Another parallel with the NOAA model is that this commons approach to collocating data is inherently light on assumptions or rules. With weather, we rely on historical data patterns to drive the predictive power of future patterns. As patterns change and new co-occurrences of events are added to the mix (for instance, we have gone from having just a few markers of weather to over 60 now), we can rely on the past to help us make unbiased predictions as to what future weather will look like.

As history has taught us, we cannot predict now the health challenges of the future. In time, however, we will solve our problems with the poor nutritional value of factory food and our deplorable sedentary behavior. What will follow? We are sure to encounter new pressures that negatively influence the health and prosperity of mankind. Clearly, rebuilding a new data architecture for future human challenges might be required if our current one was inappropriately burdened with assumption rich rules that no longer apply in future societies.

Thus, to effectively future-proof our data commons, we must resist the urge to insert or apply excessive "intelligence" that may prevent underlying truths from surfacing. Sticking to the simple NOAA model will allow us to collect co-occurrences in increasingly complex combinations, without bias, to ensure that new relationships are forever exposed through large numbers.

Author’s home page
Other articles in chronological order
Index to other articles