By Aaron Krol 

May 29, 2015 | PatientsLikeMe’s centerpiece is its social network: a site where patients living with major chronic illnesses can find other people dealing with the same symptoms to share treatment experiences, ways of coping, and research opportunities. Facebook was just two years old when PatientsLikeMe launched its first “community” in 2006, for patients with amyotrophic lateral sclerosis (ALS), yet the organization’s founders already had an intuitive sense for the strengths of social media, which would quickly make the service a popular meeting space for patients.

Key to the website’s success is its search function. Members of PatientsLikeMe communities enrich their profiles by recording symptoms they’ve experienced, along with other information like their medications and side effects. They can then search the entire site for other members who have dealt with the same or similar symptoms. These matches are the beginning of the personal contacts members forge among themselves, the core of PatientsLikeMe’s mission.

At the time PatientsLikeMe was building its platform, this kind of user-defined tagging of content was seen as a new feature of the social web, and a big departure from traditional search functions that relied on top-down data structures. Letting members enter their experiences with disease in their own words allowed PatientsLikeMe to build its entire data scheme on what the organization calls “the patient voice.”

However, says Sally Okun, PatientsLikeMe’s Vice President for Advocacy, Policy and Patient Safety, the same democratic approach to patient data that made the platform so valuable to its users would later prove to be an obstacle to connecting more formally with physicians and researchers. “The tagging was useful in a social media perspective, or a social networking perspective,” she says, “but it wasn’t going to be particularly useful in a clinical perspective.”

No one will be surprised to hear that the words patients use to describe their illnesses are imperfect matches for the terms preferred by clinicians. Take, for example, a patient who has recently developed tremors that interfere with her ability to walk. “As a clinical concept that might be considered ‘gait disturbance,’” says Okun. “And yet our patients might tell us about that in a variety of ways. They might say, ‘I was walking like a drunk,’ or, ‘I was tripping over my feet.’”

There are plenty of reasons, good and bad, for this language barrier. It’s helpful for clinicians to have highly precise terms, so it’s possible to tell at a glance when two cases have overlapping presentations. Broad analyses across hospitals or research groups also benefit from a standard vocabulary, which makes it easier for computers to reconcile data collected in different locations. And no doubt any doctor would hesitate to say a patient was “walking like a drunk” in a clinical report or scientific journal.

Yet, says Okun, the scientific community can be missing a lot more than a flair for colorful expressions when it tries to force the patient voice into a chart-shaped box. The standard scales for pain or fatigue, for instance, where patients are asked to rate their symptoms from one to ten or choose between words like “moderate” and “severe,” can obscure the patient experience as easily as clarify it.

“Even ‘severe fatigue’ does not measure the kind of fatigue many people with fibromyalgia may experience after having a particularly bad day,” Okun says. “They may have to say something like ‘overwhelming fatigue’ in order to get the actual severity that makes sense for them.” That nuance can mean the difference between aggressively treating a symptom, and failing to appreciate its debilitating effect on a patient’s life.

“The problem with these standardized terminologies is they really weren’t created to help improve care,” she says. “Many of them were created to help improve reimbursement. So they’re not tied to some true experience that a patient might be having.”

Vocabulary Building 

One of the first occasions PatientsLikeMe had to confront the messiness of its patient data concerned a vocabulary called MedDRA, the Medical Dictionary for Regulatory Activities. The PatientsLikeMe team had always hoped that its work bringing communities of patients together could be used to support clinical trials of new therapies, or extended research of treatments already on the market. To communicate with the pharmaceutical companies who would sponsor that research, PatientsLikeMe would have to learn to speak the same language.

“As we were starting to work with the pharmaceutical industry, we had an important responsibility to help them meet their regulatory obligations around adverse event reporting,” says Okun. That meant digging into the side effects that members were recording in their profiles, and squaring them with MedDRA, a formal vocabulary for adverse events that helps both industry and regulators analyze how patients are responding to drugs.

Mapping between the patient voice and a clinical taxonomy is not a straightforward task. Sometimes, as in the example of “gait disturbance,” there’s a fairly clear one-to-one correspondence: on the front end, members can still see the terms they entered, but behind the scenes those terms have been translated to a common formal taxonomy. That way, not only can a member’s data be usefully analyzed outside the PatientsLikeMe network, but when users search for one another, the patient who wrote that he’s tripping over his feet can be linked with another who said she’s always stumbling.

Other times, however, users might supply information that has no easy equivalent in a dictionary like MedDRA. Some expressions, like “cramps,” are too general; others are more specific than medical taxonomies are equipped to handle. If symptoms are felt only in a certain body part, at particular times of day, or in association with other symptoms (like “headache after coughing fits”), that information is likely to be lost in translation. These nuances can be important, both to the patients themselves and to researchers trying to understand their diseases.

Even changes in emphasis and shades of meaning can reveal real clinical differences. If two patients both rate their pain as a “five” on a numerical scale, but one describes a “hammering pain” and the other a “shooting pain,” they could be results of different pathologies. The same can be said for broader statistical analyses: fibromyalgia patients and ALS patients might score themselves similarly on fatigue scales, but if the fibromyalgia population is more likely to use adjectives like “overwhelming” or “debilitating,” that probably says something about the disease itself.

“We already know that patients will rate their symptomology differently from their clinicians,” says Okun. “If all we’re asking as clinicians is ‘tell me about your pain,’ we’re not getting in deep enough to say, ‘tell me about that icepick pain that you described.’”

Despite the translation issues, PatientsLikeMe was able to render a large number of patient-supplied symptoms in the language of MedDRA, often reaching out to members to check whether certain formal terms were accurate descriptions of their experience. At the same time, the PatientsLikeMe team decided to link their “patient taxonomy” to several other formal vocabularies that were becoming standards in electronic health records (EHRs), including SNOMED, ICD, and LOINC. “In 2008 that was a relatively forward-thinking approach that we took,” says Okun. “And I’m glad that we did, because now all of our data can map to any one of those taxonomies.”

Over the next several years, federal incentives for EHRs created with the Affordable Care Act would make these vocabularies near-universal across American hospitals and care centers. While the correspondence between member-supplied terms in PatientsLikeMe and the taxonomies in EHRs is still far from exact, PatientsLikeMe now has a fairly strong base by which to share or compare data across these systems.

Word Choice 

In its internal research, PatientsLikeMe has taken advantage of surprising patient-supplied symptoms to contribute to our understanding of chronic diseases. In the ALS community, frequent yawning turned out to be a common characteristic of one unique subtype of the disease. In the Parkinson’s community, member surveys lent support to the now well-established notion that dopamine agonists, a common class of Parkinson’s medication, can lead to addictive gambling ― and also suggested the side effect could be more widespread than previously believed.

So far, these studies that begin with hints in the PatientsLikeMe data have had to be followed up with more rigorous surveys and better definitions of the symptoms being examined. The organization is still debating what sort of work would have to be done to validate that the connections between its patient taxonomy and formal vocabularies are accurate. The PatientsLikeMe members are also in many ways a highly unusual population, and few conclusions drawn from their ranks can be safely generalized to patients as a whole.

If, however, the healthcare community can settle on a way to reliably link patient-supplied terms with clinical taxonomies, this kind of analysis could one day be applied across the healthcare system, and perhaps even automated. Physicians already collect subjective notes based on patients’ descriptions of their symptoms, and often include these in the medical record; a taxonomic system that makes even a portion of that information machine-readable would allow clinical informatics programs to draw broader conclusions about a disease’s pathology.

“The need is for us to find electronic and digital ways to start capturing this and depositing it into a more patient-focused and patient-centric data collection environment… where data’s going in and being subjected to some kind of analysis,” says Okun. This will become even more important as PatientsLikeMe starts to connect more directly with the care centers where patient data is collected. Just this week, the organization announced a partnership with the Partners HealthCare hospital network in Massachusetts, to offer patients a direct portal between the Partners Patient Gateway health management tool and the PatientsLikeMe site.

Meanwhile, as PatientsLikeMe refines its patient taxonomy to make closer matches to the language of EHRs, the creators of some formal vocabularies are interested in making their dictionaries a little more like the language of PatientsLikeMe. Okun says that MedDRA, for instance, has discussed including some of the terms that PatientsLikeMe members commonly use, to ensure its dictionary of side effects corresponds as closely as possible to drugs’ real-world impact on patients.

She is also encouraged by a recent movement in oncology to take the patient voice into account when describing symptoms associated with chemotherapy, whose effects can be hard to manage and have a huge impact on quality of life. Side effects of cancer treatments are often under-addressed, and a growing number of advocates believe that communication breakdowns between patients and their physicians are partly responsible.

Emerging technologies like natural language processing could make the transition to using the patient voice much easier in the future, opening up free text fields in both EHRs and platforms like PatientsLikeMe to much more analysis. “The time has come,” says Okun. “Patient voice, and patient-generated data broadly, has become something that people are talking about all the time.”

The focus on the patient voice is about much more than research and data collection, she adds. Even at the individual level of the doctor-patient relationship, a shared understanding of the language used to describe health can change the way care is delivered. “The goal is truly for us to better understand each other.”