home   |   research   |   blog   |   code

The layout of this page will likely change and I still want to add some figures/illustrations but I figure it is best to have something up now rather than many months from now. Also … my git-foo needs work and I was worried about losing this file in a git-merge!

My Research

Below I summarize some of the research projects that I’m currently working on or that I’ve worked on in the past few years. My training is in electrical and biomedical engineering, with a focus in neural engineering. Although my recent research has focused largely on treating urgency urinary incontinence, I’m interested in urinary function/dysfunction more generally. I also hope to branch out into other organ systems at some point, with a focus on physiological testing and diagnosis/outcomes modeling.

Motivation: Urinary Dysfunction

My current research focuses primarily on how to treat urgency urinary incontinence, or as I explain to my family, helping people who can’t make it to the bathroom in time and pee in their pants. A quote that I’ve found to be really impactful when thinking about incontinence comes from a patient that said:

“Incontinence doesn’t kill you - it just takes away your life”

(PMID: 20516418)

From the same article as that quote, urinary incontinence:

“has a profound negative impact quality of life, exceeding that of many comorbid diseases (ie, diabetes, stroke, and arthritis in the hands and wrists).30,31. Urinary incontinence is associated with a 30% increase in functional decline, and a 2-fold increased risk of falls, depressive symptoms, and nursing home placement.31-36 Further, women with UI often have concomitant and treatable symptoms that can be quite bothersome including urinary frequency, urgency, nocturia, and fecal incontinence”

Unfortunately, incontinence and other urinary symptoms are also highly prevalent. Incontinence, along with urinary urgency, urinary frequency, and nocturia (together known as overactive bladder symptoms), are estimated to occur in over 40% of women over the age of 40 in the United States (PMID: 21256571). That same article estimates the prevalence of overactive bladder symptoms in men to be greater than 25%. Men tend not to be incontinent, but rather have problems emptying their bladders and subsequently needing to go all the time (often referred to as BPH).

Regarding treatments for urinary dysfunction, there are two important things to know. First, there exists many treatments that may help. I’ve heard patients say, “I wish I had come in sooner to get treatment.” Too often people think urinary dysfunction is just a part of getting older. Although age is strongly correlated with urinary dysfunction, that doesn’t mean that urinary dysfunction is inevitable or untreatable. That being said, the second important thing to know is that there remains much room for improving treatments. That’s where I come in.

Predicting Treatment Outcomes

TODO: Insert figure …

Note: This project has been selected for 3 years of funding from the NIDDK (K01) starting in September 2020.

I’m starting a project where the goal is to predict treatment outcomes for two 3rd line therapies for treating overactive bladder, Botox (administered directly into the bladder) and sacral neuromodulation. These are two treatment options for those that fail medication therapy. On average these two treatments are about equally effective. However, we don’t know if some patients are really good candidates for one of those therapies but not the other. Similarly, we don’t know if some patients have almost no chance of responding to these two therapies, and would thus be a good candidate for other novel therapies. This project involves using predictive modeling/machine learning approaches to try and answer whether someone will have a successful response to a therapy, or whether they will experience an adverse event, based on pre-treatment measurements.

This project is really exciting to me for many reasons. First, predictive modeling and machine learning are useful ways of answering important questions, (e.g., will my patient respond to this treatment). I think these approaches are underutilized in the field, and I hope to be one of the people pushing these techniques. Second, these techniques are good for analyzing multi-dimensional data. A lot of research in the field (and I’m guessing elsewhere) focuses on single factors that determine symptoms. However, many analyses show that single factors fail to explain/separate those that have symptoms and those that don’t. In other words, it’s often not one single thing that contributes to a disease or symptoms, but a variety of things in combination. Finally, using these approaches we can begin to understand who is not being well treated by current therapies. This is extremely useful information when looking to develop new therapies.

Organ-Based Phenotyping

TODO: Insert figure

I am currently participating in a NIDDK sponsored network study called Symptoms of Lower Urinary Tract Dysfunction Research Network or LURN. The primary goal of LURN is to better understand “types” of patients (i.e., phenotyping), with the expectation that improving our understanding of patients will lead to improved treatments. LURN has multiple sub-groups that are using different approaches to phenotype patients. The group I’m working with is called LURN-organ, which is focusing on how we can phenotype patients based on physiological, organ-based, testing.

We, LURN-organ, designed a testing protocol for taking physiological measurements of the bladder and urethra from women with urgency urinary incontinence, urinary urgency without incontinence, and asymptomatic controls. Clinical testing at six-sites is scheduled to start in mid to late 2020. I’m excited as clinical trials tend to be focused on evaluating a therapy (e.g., does this drug work), and not on physiological testing to better understand patients. Instead, physiological testing tends to be done at single sites, and thus generally with a small group of patients. Additionally, authors of these studies tend not to relate these results to treatment outcomes. Finally, the patient populations are often heterogenous (which is fine), but not when the patient population is poorly characterized and small. This makes it difficult to relate the results of the studies to patients that a clinician sees. The end result is that these studies, in my opinion, tend to be ignored in the long run. This is unfortunate as I believe a better grasp of the pathophysiology will really help with treatment selection and design of new therapies. Similar to the clinical predictive modeling/machine learning mentioned above, I see tremendous opportunities in this area.

On-Demand Prevention of Incontinence

TODO: Insert figure

Studies have suggested that in some people, just before they have an incontinence episode, their urethra relaxes inappropriately. This is similar to what we would expect if someone were volitionally voiding. Although this has been known for decades no therapy directly addresses this observation.

In many cases of incontinence people have some advanced warning, they just are not able to make it to the bathroom in time. I hypothesized that it may be possible to electrically stimulate the urethra to either cause it to contract, or simply to prevent it from relaxing, while ambulating to the bathroom to prevent incontinence. I conducted a feasibility study in eleven patients to determine if a particular approach to stimulating the urethra, intraurethral stimulation, showed any evidence of elevating urethral pressures. This approach is desirable because it is relatively straightforward to do and minimally invasive compared to other approaches. Unfortunately, the intraurethral stimulation approach did not work, but I think we learned some interesting things along the way, and the study is an important first step in pursuing this novel therapy. I’m currently revising a paper that summarizes this work based on reviewer comments.

Development of this novel therapy also highlights the importance of 1) improved physiological testing to understand patients that exhibit variations of this particular phenotype and 2) analysis of which of these patients, if any, are already well treated using current therapies.

Pudendal nerve stimulation

TODO: Insert figure

There are currently two electrical stimulation therapies that are used clinically to treat urinary dysfunction, stimulation of sacral nerves or the tibial nerve. The tibial nerve is stimulated by placing an electrode close to the ankle. Sacral nerve stimulation involves stimulating nerves close to the point where they exit the spinal cord. There are however other locations that stimulation of the nervous system can occur, such as the spinal cord, in the brain, at nerves closer to the target organs (bladder, urethra), or even external stimulation of the genitalia.

One nerve that may be promising to stimulate electrically is the pudendal nerve. The pudendal nerve forms from sacral nerves and eventually innervates the urethra and genitalia. I’ve conducted numerous studies in animal models trying to characterize pudendal nerve stimulation and exploring ways to make pudendal nerve stimulation more effective.

I’m currently not doing any pudendal nerve stimulation work, but I’ve been thinking a lot about what studies are necessary to really move this therapy into the clinic. Interestingly, pudendal nerve stimulation has been used clinically, using a sacral nerve stimulator placed at the pudendal nerve. In 2005 a blinded randomized study of 30 patients was done comparing pudendal nerve stimulation to sacral nerve stimulation. Patients received 7 days of stimulation at one location then 7 days of stimulation at the other location. Out of the 24 patients that responded to either stimulation, 19 preferred pudendal nerve stimulation (PMID: 16178000). I have my own pet theories as to why pudendal nerve stimulation may be superior to sacral nerve stimulation in some cases, but further testing needs to be done. As no company has a FDA approved device that is really pushing for pudendal nerve stimulation, it remains relatively unused in clinical practice.

You can read about my initial work with pudendal nerve stimulation here as well as similar work we’ve done with stimulation of another nerve, the pelvic nerve, here. Three more followup papers will hopefully be coming out soon related to peripheral nerve stimulation to treat bladder dysfunction.

Intravesical Prostaglandin E2

TODO: Insert figure

When doing animal research it is important to have animal models that recreate some aspect of the disease you are studying. When I first started doing urological research my advisor wanted to investigate placing Prostaglandin E2 (PGE2) in the bladder of rats as an animal model of overactive bladder symptoms. Interestingly, PGE2 has been placed in the bladder of healthy women, and it led to urinary urgency, the main symptom of overactive bladder.

Initially, the plan was to simply use intravesical PGE2 (i.e. PGE2 placed in the bladder) to reduce the amount of fluid that animals could hold in their bladder before they reflexively emptied their bladders. Then we would follow it up with different types of electrical stimulation to see which types worked best. Along the way I made some interesting observations regarding PGE2 that hadn’t been well described before. These observations have reinforced in my mind a cause of urinary urgency that although it may not be completely novel (rediscovered?) is definitely under appreciated in the field. As urinary urgency is poorly understood, it is valuable to gain additional insights into factors that may cause it. This has also generated some ideas as to how we might test for this type of urinary urgency and therapies that might be created to treat it.

You can read more about this here. I’ve also written a follow-up paper that we’re currently editing and that will be submitted for publication soon. The follow-up paper was presented at a NIDDK meeting in December 2019 and won a best-poster award.

Older Projects

The following projects are slightly older but still relevant in terms of my research interests.

Electrical Stimulation Modeling

TODO: Insert figure

In graduate school we were examining whether scaling the instantaneous stimulation rate on a small set of electrodes (n=16) placed in dorsal root ganglia would lead to brain responses that more closely resembled the activity we saw when we manually moved the animal’s limb. Nerves tend to have lots of cells (neurons) inside of them (tens of thousands?). Currently, electrical stimulation devices tends to be limited to at most 10s of channels, meaning we can’t stimulate each neural cell individually. We wanted to examine if we could make up for the lack of stimulation channels by “stimulating harder” (i.e. increasing the rate of stimulation) on the electrodes we were using.

The overall goal of this work was to study how to best deliver artificial somatosensory feedback for prosthetic users. Put another way, we - and others - were investigating ways of stimulating the nervous system so that touching or moving a prosthetic limb delivered sensory feedback to the user. So for example if someone with a prosthetic limb touched something, electrical stimulation would stimulate the person’s nervous system in such a way that the person would “feel” whatever they were touching.

Interestingly, when I scaled the instantaneous stimulation rate we started to get really reliable brain responses that were locked to the co-occurrence of stimulation on neighboring stimulation channels. In other words, if two channels that were physically close to each other on the electrode array stimulated at overlapping times we were likely to see brain activity that was evoked from those stimuli. However, if those pulses were shifted slightly in time so that they no longer were overlapping, this effect went away.

Using computational modeling I was able to show that simultaneous stimulation on nearby electrodes activated a lot more neural tissue than just the sum of the tissue activated when the two electrodes stimulated independently. This was known for very large electrodes stimulating on the outside of nerves, but for small electrodes stimulating inside of nerves at low stimulus amplitudes, the electrodes were thought to be independent. The end result of this work is that if these types of microelectrodes are used, we now know that most likely they will need to stimulate on each electrode at slightly different times to avoid interactions between the channels. Alternatively, it should be possible to stimulate on multiple electrodes at the same time to get unique activation of neural tissue.

You can read more about this work here.

Worm Modeling

TODO: Insert figure

In graduate school I wanted to learn how to program in Python. In general, I find it easier to learn a programming language when I’m using it to accomplish something. At the same time a friend told me about a group, OpenWorm, that had formed online that was trying to do open, collaborative, scientific work centered around trying to accurately model a worm, the C. Elegans.

C. Elegans are interesting for a couple of reasons. Surprisingly, C. Elegans share enough genetic information with humans that we can learn a lot about human genetics by studying C. Elegans. Second, the connections of each neuron to every other neuron, known as the connectome, have been mapped. C. Elegans only have only 302 neurons whereas humans have tens of billions. OpenWorm is interested in seeing how closely we can come to modeling a C. Elegans by using the wealth of information that scientists have already collected about the worm, including notably the worm’s connectome. The hope is that by creating an ultra-realistic model of a C. Elegans that it will help guide and inform learning about more complex organisms.

During my time with OpenWorm I focused on translating code from Matlab to Python that quantified how worms moved, such as how fast they moved, how long they would stop moving, how often they changed direction, etc. By quantifying as many parameters as possible, other scientists could use that information to tease out differences in behavior between different genetic variants of the worm. This could ultimately be used to improve our understanding of human genes. For OpenWorm, this also gave us a reference that we could compare the behavior of our simulated worms to. Differences between real and simulated worms would indicate that our simulations needed to be improved.

This project also instilled in me the notion of trying to infer differences in function based on quantifying as much as possible from time-series data. Time-series data, basically something sampled over time like a bladder pressure trace, the x-y positions of a worm, or an audio recording, have the potential to be extremely feature rich. In other words, rather than just calculating something like the mean (e.g., this audio recording is quiet or loud), local patterns can be extracted to give us more information (e.g., there’s a 5 minute drum solo, the recording contains brass instruments, their are 3 verses, etc.). I’m planning on applying this type of analysis to the time-series data that are routinely collected in urology, and where currently relatively simple analyses tend to be performed. The hope is that this more in depth analysis tells us more about a patient so that we can improve their treatment.

You can read more about OpenWorm here and our work on analyzing worm movementhere which was based on this work.