One strategy for medical innovation is to ask: “What’s the feedback signal?”

How do you know whether a treatment is any good? Traditionally, there have been two options to answer that: First, rely on your own impression as patient or physician, and second, rely on data from clinical studies.

1. Relying on impression is problematic because both parties are biased. You want to be better, and your doctor wants you to be better. So even if the pill did absolutely nothing, you are likely to conclude that it helped at least a bit. Phenomena such as regression to the mean and the asymmetric doctor-patient relationship are the reason why it’s hard to figure out what works and what doesn’t, and therefore we are still having exactly the same debates about homeopathy and other quackery as 200 years ago.
2. Clinical studies on the other hand take results from a controlled and double blinded representative group and extrapolate them to a defined patient population at large. While this method is standard practice in drug development for good reason, it is not immune to attempts of manipulation. But at least there are solid ways to estimate the quality of any published evidence. For some surprisingly entertaining stories from this world, I suggest reading and watching Ben Goldacre.

Feedback controlled therapies

There is a third option, more personalised and based on the ease and acceptance of collecting and mining practically anything that can be measured. The disruptive potential doesn’t come from biomarker-guided therapy as such, but from minimising the cost and effort it takes to obtain as many data points as possible.

It is crucial to understand what exactly can and should be measured, and what to do with those data. Often, the thing that can be measured (a surrogate parameter) is not the same as the thing we care about (a clinical outcome). For any bio-signal, it needs to be understood how well it actually represents or predicts the thing we want to monitor.

So here is an assorted and certainly incomplete list of a few developments:

Diabetes

The most elegant example for what you could call feedback controlled therapy comes from diabetes. It is now possible to have a continuous closed feedback loop of glucose monitoring and insulin application. Because blood glucose is such a good predictor of long term sequelae of diabetes, it’s sufficient to keep that one value within its limits to reduce the risk of stroke, kidney failure, blindness or polyneuropathy. Take this principle of closed-loop control, and ask where else it could be applied.

Blood Pressure

Blood pressure is equally straightforward as a predictive parameter, but much harder to turn into an actionable stream of data. Home blood pressure monitors involve too much active handling to really count, and invasive arterial catheterisation is confined to intensive care. But what if it’s possible to come up with a wearable device for continuous blood pressure measurement, that isn’t invasive and doesn’t require a cuff? It wouldn’t even have to be as precise as the best current devices — it just needs to be good enough; the superiority comes from the quantity of data points. There are plenty of choices of effective antihypertensive drugs, and yet high blood pressure is still a silent killer. The solution clearly isn’t new drugs or drug combinations, and it’s also not a hypertension app or diary storing three values a day. The solution is a proper continuous feedback signal, so that the disease is no longer under-monitored and under-treated. A quick search brings up e.g. this open access paper, which shows that continuous blood pressure measurement on a wristwatch is definitely in the realm of possibilities.

Heart disease

It is worth mentioning that ECG monitoring with an Apple Watch seemed like science fiction not long ago. This function is currently limited to the detection of atrial fibrillation, which is the most common type of cardiac arrhythmia. In order to obtain an ECG, you need to place a finger on the crown button of the watch and wait for 30 seconds. Since atrial fibrillation can be paroxysmal and asymptomatic, the Apple Watch doesn’t replace a Holter ECG or loop recorder. Loop recorders are impressive devices by themselves, basically a continuous ECG measurement capsule placed under the skin of the chest. It’s conceivable that smart loop recorders could work synergistically with other devices, and they will definitely be used more frequently in cardiac patients.

To give a different, well established example from Cardiology, an Implantable Cardioverter-Defibrillator is basically a closed-loop immediate treatment device acting upon detection of a pathological heart signal.

Lungpass

Lungpass is a promising start-up, with a dedicated CEO, that aims to automatically analyse auscultation sounds. They have done the validation to confirm that their machine learning method of classifying lung auscultation phenomena can keep up with expert examiners, with pretty good sensitivity and specificity of generally well above 80%, for normal sounds, wheezes, rhonchi and crackles. The principle, again, is taking an established but manual measurement device, in this case a stethoscope, and turn it into an actionable data stream.

Inflammatory Bowel Disease

The most relevant current biomarker here is calprotectin, an antimicrobial, metal-chelating protein complex secreted by neutrophils, that is measured in stool by many laboratories as part of regular doctor check-up visits. While the idea of home monitoring of inflammatory activity for patients with IBD is good in principle, there are definitely still some road blocks ahead, as you can probably imagine if you’re able to endure this video. Even if you would regularly and correctly use this laborious, time consuming kit, and trust the validation studies of the company, it just doesn’t give you a lot of data points; probably in the order of one a month.

Maybe this is enough for a biomarker-guided therapy, and everything else would be overkill. But who knows, somebody might come up with a functioning wearable patch just like a glucose monitor, that delivers a continuous inflammation signal, and we’ll realise that it could be useful for all kinds of other chronic inflammatory conditions as well, and we could ask whether close monitoring has steroid-sparing potential.

Theranos

Speaking of wearable patches. Theranos was a massively hyped blood testing start-up that fraudulently overpromised, indicating how much pressure is on for finding cheaper, lab-independent monitoring methods of blood values. The peak valuation of the company in 2014 was at $10 billion, and it eventually stopped operations in 2018. But it’s worth recalling the initial idea, which was a wearable patch that could inform drug delivery adjustments and notify doctors of patients’ blood test values. The $10 billion hype was essentially about the prospect of closing multiple feedback loops.

To give just one example, warfarin blood thinning therapy needs to be closely monitored by measuring the INR from a drop of blood, because the actual drug effect fluctuates widely between individuals and depending on dietary habits. So, a wearable patch could tell your phone to tell you to take x amount of warfarin for your next dose. Clearly, this requires the test results to be absolutely reliable, which Theranos could not deliver, posing a considerable safety issue.

At least the whole thing makes a good story for a book, with a film version featuring Jennifer Lawrence apparently in production.

Movement Disorders

This one is tricky, and would deserve an article on its own, because despite an endless choice of accelerometers out there, Movement Disorders are not an easy application. The evaluation of pathological movements, e.g. in patients with Parkinson’s Disease (PD), classically rests on observation, and actually touching patients. One of the hallmark features of PD is rigidity, a characteristic lead-pipe like resistance to passive joint movement, that has a reputation of being impossible to measure with a machine. The only way to assess rigidity is to take a patients’ arm with both of your hands and passively move it, and then grade your subjective impression of symptom severity according to a clinical scale.

So are there any surrogate parameters that could inform you about the severity of rigidity? Not really, but decreased or increased movements can be tracked. At the moment, the product that’s closest to reliably logging movement ability in PD patients is the PKG system, a wrist device in dire need of a better design, that provides a graphical overview of bradykinesia, dyskinesia and medication times.

Given that the current standard management of PD consists of approximately half-yearly doctor appointments providing only a snapshot assessment, and medication for symptoms that can wildly fluctuate within a day, it becomes clear that there is a potential for improving feedback control here.

Recap

Actively looking out for diseases that currently have insufficient feedback signals, i.e. slow/few/expensive methods for measurement and monitoring, can point towards areas of potential medical innovation. However, it’s important to remain critical about whether the new fancy technology is actually better than established low-tech or no-tech alternatives.

For example, patients with Inflammatory Bowel Disease can usually tell when their disease may be flaring because of symptoms like abdominal pain and diarrhoea, and regardless of a test kit result, they are likely going to see their treating physician anyways. It’s like smart technology to detect spoiled milk: Your nose is going to tell you that. The only way this makes sense, is if a flare-up can be anticipated by a few days.

Similarly, patients with Chronic Obstructive Pulmonary Disease might not need machine learning auscultation analysis at a time point when they have already noticed heavier breathing and increased sputum production; maybe a cheap plastic peak flow meter suffices for that. If a new marker is sensitive enough to predict an exacerbation, it becomes a different story.

The most important criterion is, can a feedback signal be obtained that requires as little input as possible from the user? Symptoms will be overlooked, and devices that need attention will not get used. So what’s needed are monitors in the background that only demand attention when something needs to be done. Wherever that is fulfilled, there is a potential innovation.