index.html (14110B)
1 <!DOCTYPE html> 2 <html lang="en"> 3 <head> 4 <link rel="stylesheet" href="/style.css" type="text/css"> 5 <meta charset="utf-8"> 6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> 7 <meta name="viewport" content="width=device-width, initial-scale=1.0"> 8 <link rel="stylesheet" type="text/css" href="/style.css"> 9 <link rel="icon" href="data:image/svg+xml,<svg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 100 100%22><text y=%22.9em%22 font-size=%2290%22>🏕️</text></svg>"> 10 <title></title> 11 </head> 12 <body> 13 <div id="page-wrapper"> 14 <div id="header" role="banner"> 15 <header class="banner"> 16 <div id="banner-text"> 17 <span class="banner-title"><a href="/">beauhilton</a></span> 18 </div> 19 </header> 20 <nav> 21 <a class="nav-active" href="/about">about</a> 22 <a href="/now">now</a> 23 <a href="/thanks">thanks</a> 24 <a href="/posts">posts</a> 25 <a href="https://notes.beauhilton.com">notes</a> 26 <a href="https://talks.beauhilton.com">talks</a> 27 <a href="https://git.beauhilton.com">git</a> 28 <a href="/contact">contact</a> 29 <a href="/atom.xml">rss</a> 30 </nav> 31 </div> 32 <main> 33 <h2> 34 husband and father 35 </h2> 36 <p> 37 Talented, beautiful people surround me. The <a href="/now">now 38 page</a> usually has the most current updates in this arena. 39 </p> 40 <h2> 41 physician 42 </h2> 43 <p> 44 Oncology fellow, Vanderbilt 2022-2025.<br> 45 Internal medicine resident, Vanderbilt 2020-2022.<br> 46 Medical school, Cleveland Clinic Lerner College of Medicine of Case 47 Western Reserve University 2015-2020. 48 </p> 49 <p> 50 I enjoy treating people with any cancer, but have a particular 51 affinity for head-and-neck and upper GI cancers (together, 2/3 of the 52 “aerodigestive” cancers, the other 1/3 being lung). There are many right 53 ways to be a doctor, but the way I understand my role is this: to be 54 with people in their suffering and their joy (suffering is everywhere, 55 but there is a special joy known only in the oncology clinic), to see 56 and know the soul within the body (whatever the tumor or the scalpel or 57 the radiation beam or the drug has done to that body), to cure when I 58 can (and deal with the fallout from that cure), and help always (to the 59 end and beyond). 60 </p> 61 <p> 62 There are glimmers of hope and serious advances in recent years, but 63 these remain horrible diseases, many of which have few treatment options 64 once they’ve reached an advanced stage, and the treatments we do have 65 tend to be quite difficult to tolerate. Most cancers can be disfiguring, 66 but these especially so, whether from the tumors themselves or from the 67 therapies. Aerodigestive cancers also disproportionately affect folks 68 who live out in the country, or in the cities but with few resources, 69 and these folks are my folks. 70 </p> 71 <p> 72 We need to help each other, and there is much to do. 73 </p> 74 <h2> 75 educator 76 </h2> 77 <p> 78 Harvard Macy Institute faculty, 2018-2020. Health Care Education 79 2.0. 80 </p> 81 <p> 82 <a href="https://vimbook.org">https://vimbook.org</a>, 83 2020-present.<br> 84 The Vanderbilt Internal Medicine Handbook was started by Mike Neuss, 85 MD/PhD in the late 2010s when he was a resident. It’s an incredible 86 resource, primarily envisioned as a physical book to keep in your white 87 coat pocket for quick, authoritative reference. It had a website when I 88 came to Vandy, but the UX… left something to be desired. I rebuilt it 89 into its current state (website and infrastructure only - each section 90 has its own author(s)), and help maintain the back end. 91 </p> 92 <p> 93 Chase Webber is the faculty support, and has been amazing. 94 </p> 95 <p> 96 It’s used globally, and one of my goals is to make it easier for 97 smaller, particularly international programs to have their own versions. 98 It uses only free and open-source software, and fits comfortably into 99 pretty much any free server despite being a fairly large book with 100 multimedia. 101 </p> 102 <p> 103 The main problem is that updating the content, while it is just in 104 Markdown (a very simple text format you can learn in ~7 minutes), feels 105 a little too tech-y to be comfortable for people who haven’t spent time 106 in a code text editor. What it really needs is funding, and a part-time 107 developer with protected time. 108 </p> 109 <h2> 110 data scientist 111 </h2> 112 <p> 113 <a href="https://scholar.google.com/citations?user=Ng5AgXAAAAAJ">Google 114 Scholar profile</a> 115 </p> 116 <p> 117 <em>If you are interested data science consulting, <a href="/contact">contact me</a>. Current rate is listed at that 118 link.</em> 119 </p> 120 <p> 121 <em>Major projects are listed below in reverse chronological order 122 (roughly), newest projects at the top.</em> 123 </p> 124 <h3> 125 overview 126 </h3> 127 <p> 128 My work has morphed over the years, as everyone’s does, but the 129 consistent thread and drive throughout has been on coaxing large, messy, 130 complex data to tell us a story about ourselves, about all of us as 131 societies and neighborhoods as well as each one of us individually, to 132 empower us to speak our own sequels. 133 </p> 134 <p> 135 To translate that into buzzwords, for your bingo game: I’m an 136 oncologist who uses explainable artificial intelligence among other 137 techniques to diagnose and address healthcare disparities, including 138 democratizing personalized medicine and pursuing synergies in the global 139 academic-industrial complex. 140 </p> 141 <p> 142 I was trained as an anthropologist, where reductionism is an insult, 143 at the same time that I was trained as a scientist, where reductionism 144 is the central conceit. Despite the way I set up that last sentence, 145 there is no true conflict, as the goal for most scientists and 146 anthropologists is the same: to make things at least a little better for 147 someone, but hopefully a lot better for everyone. Modern anthropology is 148 inherently activist, far from the crusty sepia-toned image of a staid 149 researcher sitting just outside the village campfire furiously 150 scribbling in a notebook, and so is modern medical research (we all read 151 Tuskegee and are appalled and want to do better, though precious few 152 become <a href="https://en.wikipedia.org/wiki/Paul_Farmer">Paul 153 Farmer</a>). 154 </p> 155 <p> 156 Data science was an unexpected boon, a set of tools that lets me deal 157 with staggering complexity in a disciplined way, to a degree unifying 158 the anthropologist’s drive to let the data be itself (messy, human) with 159 the scientist’s drive to simplify. 160 </p> 161 <p> 162 I became acquainted with using code to model the world when I was 163 doing physics research (equivalent circuit modeling and scanning laser 164 doppler vibrometry on a Nigerian-style clay pot drum, total hoot, I <a href="https://doi.org/10.1121/1.3654998">presented</a> it at the 165 Acoustical Society of America Annual Meeting and we published it in <a href="https://doi.org/10.1121/1.4789892">JASA</a>), and the 166 research-centric medical school at Cleveland Clinic gave me the better 167 part of five years and a supportive environment to dive in deeper to 168 computer science, machine learning, etc. 169 </p> 170 <p> 171 Now that have the freedom to choose my own direction, data science 172 projects that were informed by anthropology but had centroids in other 173 disciplines have become squarely within the overlap of my personal Venn 174 diagram. 175 </p> 176 <p> 177 Additionally, as I have moved from academia to the worlds of private 178 practice and industry, I am increasingly inspired toward using these 179 interests to hit cancer where it needs it the most: in the clinic of the 180 community oncologist, where the vast majority of cancer patients receive 181 care, by bringing timely insights from every relevant axis to bear on 182 every visit. 183 </p> 184 <p> 185 <a href="https://en.wikipedia.org/wiki/Information_wants_to_be_free">Information 186 almost wants to be free</a>, and there is now so much of it, that a lot 187 can be done for free or at a low cost. (I’m take special glee in 188 byproducts, “digital waste,” some say, but there is gold in them there 189 hills - <a href="https://en.wikipedia.org/wiki/Shigeru_Ban">Shigeru 190 Ban</a> is my hero). There are abstracts from conferences aplenty 191 describing the cutting edge of everything, usually seen as jots and 192 tiddles on the march of science forward but forgotten when the journal 193 article gets published, and lab values and imaging results and clinic 194 notes generated in the usual course of patient care, used once or a few 195 times and then forgotten and discarded in the wake of what comes next - 196 in all of this digital refuse I believe there are insights to be gained, 197 efficiencies to be achieved, lives to be saved. 198 </p> 199 <hr> 200 <h3> 201 disparities in cancer research across space and time 202 </h3> 203 <p> 204 This ongoing project starts with a custom metadata archive of all the 205 abstracts from the two big annual cancer meetings (ASH and ASCO - as a 206 starting point), made machine-readable and hence more easily accessible 207 en masse. Here’s a <a href="https://ash-abstracts.vercel.app/abstracts_small/abstracts">rough 208 prelim</a> of what that looks like, with an interactive map and 209 sorting/search/export (you can hit JSON or CSV endpoints trivially to 210 get nicely formatted data, as opposed to the hulking web scrapers and 211 HTML wranglers I wrote to get the data the first time). Analyses are 212 planned on the authors and affiliations (global mapping, with a time 213 component), and how those overlap with disease states (e.g. how much 214 sickle cell research comes out of Sub-Saharan Africa? We know it’s bad, 215 but nobody has quantified how bad - the idea is to have a platform for 216 answering these questions easily), before branching out into things like 217 conflicts of interest and various analyses using NLP, etc. 218 </p> 219 <hr> 220 <h3> 221 Cleveland Clinic Center for Clinical Artificial Intelligence 222 </h3> 223 <h3> 224 explainable machine learning for readmissions, with a focus on 225 sociodemographics 226 </h3> 227 <p> 228 While at Cleveland Clinic, I was one of the founding members of the 229 Center for Clinical Artificial Intelligence, and its first dedicated 230 analyst. We had the delightful opportunity to work with a rich dataset 231 from one of the world’s largest hospitals, focused on predicting the 232 risk of readmission (discharging from the hospital and “bouncing back” 233 too soon) and extended length of stay. We had access to not only health 234 data, but socioeconomic data, for millions of patients, and I further 235 enriched this with census data. 236 </p> 237 <p> 238 With only an address and a date, one can learn about a person’s 239 neighborhood in incredible detail (though not the person themselves - 240 the US Census is wise about privacy in the data they publish). It turns 241 out that a person’s neighborhood is a major predictor in their health 242 outcomes, as is their insurance provider, in addition to a host of 243 health parameters and hospital process clues. 244 </p> 245 <p> 246 I used interpretable machine learning techniques, with a focus on 247 revealing actionable items that the patients and medical teams could 248 hope to influence, as well as to call out structural issues that 249 organizations and governments need to know. One of the problems with 250 machine learning is that it can only ever restate (and in some cases, as 251 it feeds back on itself, more firmly entrench) the biases that led to 252 the historical data you fed to it. If you design it to show you these 253 biases, show you <em>all</em> of its biases, explicitly, to tell you 254 exactly how much a person’s race, ethnicity, gender, and neighborhood 255 played a role in its predictions (alongside medical diagnoses and lab 256 values), you have transformed an algorithm from a potentially 257 destructive tool, a heinous thing that pushes hurting people farther 258 down, into a tool for positive change. If these biases are shown in an 259 easy-to-read visual, that patients, clinicians, and administrators can 260 all understand, all the better. See what it looks like <a href="https://doi.org/10.1038/s41746-020-0249-z">here</a>. 261 </p> 262 <hr> 263 <h3> 264 Cleveland Clinic Taussig Cancer Institute 265 </h3> 266 <h3> 267 explainable machine learning for blood cancer diagnostics 268 </h3> 269 <p> 270 Initially, when I was going to be a malignant hematologist (I still 271 think it is a beautiful field, but was drawn away to other pastures), I 272 wanted to level the playing field for advanced diagnostics. 273 </p> 274 <p> 275 A world-class hematopathologist together with a world-class 276 clinician, preferably a whole group of these together, are required to 277 make certain diagnoses, and these only after at least one invasive 278 biopsy. What if we could mobilize the rich genetic and phenotypic data 279 available in simple blood samples and from the electronic medical record 280 to support diagnostics, and, eventually, democratize them? Further, what 281 if the answer the machine gives could be not only accurate, but 282 interpretable? 283 </p> 284 <p> 285 What if we could get the machine to explain itself and its reasoning, 286 both as a check against biological implausibility (and more insidious 287 problems such as systemic racism), and to reveal more areas for 288 research? What if a high resolution bone marrow biopsy image could be 289 read in moments by a smart phone in rural 290 Arkansas/Kandahar/Mogadishu/Kushalnagar/etc., instead of having to be 291 shipped to one of a handful of academic centers while the patient waits, 292 still sick, for an answer? 293 </p> 294 <p> 295 We made <a href="https://doi.org/10.1182/blood-2019-126967">some</a> 296 <a href="https://doi.org/10.1182/bloodadvances.2021004755">progress</a>, 297 but these problems remain largely unsolved. 298 </p> 299 <hr> 300 <h2> 301 internal links 302 </h2> 303 <p> 304 If you want to know which tools I use, visit <a href="/uses">/uses</a>. 305 </p> 306 <p> 307 If you are interested in the tech stack for this website, visit <a href="/colophon">/colophon</a>. 308 </p> 309 <p> 310 If you would like to throw money at me, for whatever reason, visit <a href="/pay">/pay</a>. 311 </p> 312 </main> 313 <div id="footnotes"></div> 314 <footer></footer> 315 </div> 316 </body> 317 </html>