A Look Ahead

Download PDF

The end of the year is a fitting time to take stock of recent clinical developments. EyeNet asked three of its editorial board members to review their areas of expertise and to consider recent trends and news that have the greatest potential to shape their sub­specialty over the next several years. Michael E. Snyder, MD, discusses refractive-cataract pathologic dysphotopsias. Patricia Chévez-Barrios, MD, shares revolutionary thoughts on the near future of ocular oncology. And Cecilia S. Lee, MD, and two of her colleagues—Thomas Hwang, MD, and Renee C. Bovelle, MD—discuss the elephant in the room: ChatGPT.

REFRACTIVE-CATARACT

Pathologic Dysphotopsias and Thinking About Optics Differently

Michael E. Snyder, MD

Sad to say, but dysphotopsias come in many varieties. Classically we think about positive (shimmering and flickering lights) and negative dysphotopsias (typically temporal dark arcs and cresecents) as phenomena arising after cataract surgery in spite of an anatomically normal appear­ing eye and a perfectly centered IOL, but there are a host of unwanted optical phenomena that can occur when things aren’t that perfect. These unwanted phenomena can vex patients and their ophthalmologists alike. The origins are legion and not always straightforward to identify. Even when identified, some dysphotopsias are not always straightforward to treat. Accordingly, through frustration, sometimes we invoke magical think­ing and hope they will just go away. If we (and our patients) are lucky, neuroadaptation will exorcise the dysphotopsias from consciousness, yet for many, this ostrich strategy is ineffective. For the purposes of this article, we will consider only optical dysphotopsias and will defer retinal, optic nerve, and higher-order processing issues—like stroke, Charles Bonnet Syndrome, and fast flicker-fusion processors—for some other time.

The Snellen Trap

In 1862 the vision chart using letters to assess the visual angle was introduced by Dutch ophthalmol­ogist Herman Snellen. This ubiquitous black-on-white test has enjoyed over 160 years of monopo­listic dominance. Yet, it measures only one small facet of vision: high-contrast spatial resolution. And even at that, we all recognize that some letters are “easier” than others. Traditional Snellen acuity ignores contrast, color, aberration, and distortion and, thus, really has little place in a discourse about dysphotopsias. Although some negative dysphotopsias can be documented with a visual field, for pathological dysphotopsias and other vi­sual symptoms we must rely on the patient history. Dysphotopsias are often the culprit in spanning the not-so-occasional gap between “20/20” and “20/ happy.” We need to listen to what our patients are telling us, then distill that information into what makes optical sense. The following paragraphs are devoted to this process.

Where Do Pathologic Dysphotopsias Come From?

Pathologic dysphotopsias have many origins, indeed. For convenience, let’s break down the myriad of causes into those stemming from the hardware (IOLs and other implants) and the eyeball itself, recognizing that this is an arbitrary distinction in that some dysphotopsias occur because of interactions between a pseudophakos and the native anatomy.

PINHOLE TREATMENT. A Morcher pinhole implant (1A) in situ in an eye with a corneal scar and an indwelling HumanOptics custom, flexible iris prosthesis. (1B) Unfolding in the ciliary sulcus in a patient with post-RK irregular astigmatism.

Pseudophakic Dysphotopsias

Edge effects. Light striking the edge of an IOL is a common cause of positive dysphotopsias, with arcs and halos being the most widely recognized. Although there is no correlation between IOL tilt and decentration and classical dysphotopsias, an exposed edge from subluxation or iris defects can contribute to the myriad visual symptoms of pathologic dysphotopsia. The light may strike the edge either from externally or after having passed, in part, through the optic, for tangentially directed rays. Monocular diplopia or shadow images are also common in this setting. Sometimes an “en face” examination at the slit lamp can betray an exposed edge if oblique viewing is omitted from a more cursory exam, especially when there is a deep ciliary sulcus and thus ample space for light to get to the edge of the IOL from tangen­tial illumination. When an exposed edge is the culprit, IOL reposition or exchange, iris repair, or iris prosthesis use may remediate the underlying anatomy. Certainly, edge shape can impact the relative intensity of these phenomena. Rounding or frosting an IOL edge can limit these symptoms.

Optic effects. Several undesired phenomena can result from the optic of an IOL. Obviously, deep scratches on the optic that induce glare can be remedied by implant exchange. The surgeon should carefully examine presumed scratches on the posterior surface because streaks of “lubricious substance,” the proprietary slippery material used to line IOL injector cartridges, can sometimes be deposited on the back of the IOL during passage through the injector, mimicking a scratch. These streaks are a little thicker than typical scratches and often have an iridescent appearance. These can usually be easily polished off the back of an implant with a metal irrigation/aspiration tip or other smooth instrument.

Rarely, patients will notice symptoms from internal reflections within an IOL’s material. This correlates positively with the relative refractive index of the IOL.

With multifocal IOLs the optic design is in­tended to create multiple focal planes to achieve a presbyopia-correcting effect. Although some patients do not tolerate the aberrations intention­ally induced by the design of a lens, the over­whelming majority of patients with these implants do remarkably well. In these cases, the surprising fact is the relatively low incidence of unwanted aberrations, underscoring that optics is only the first component of visual processing, with image creation in tertiary cortical neural processing taking a more important role.

Manipulation of sphericity. Over the last two decades, much discussion has entered our profes­sion about the relative merits of IOLs with either neutral spherical aberration or negative asphericity to ostensibly counter the effects of the positive asphericity in the average cornea. We have all latched onto these principles, which make lots of sense when thinking about wavefronts. We have also tweaked the concept to seek out IOLs with higher negative asphericity for more oblate cor­neas and purposefully selected traditional spheri­cal IOLs for hyper-prolate post-hyperopic laser vision correction patients. Interestingly, despite the long time since introduction and wide adop­tion of these lenses, there is an absence of convinc­ing evidence to support that these machinations make any clinical difference. In my complex IOL referral practice, I have yet to see a patient present with complaints attributable to spherical aberration from a “standard” spherical IOL.

Some IOLs have tried to augment pseudopha­kic depth of focus using purposefully induced high asphericity instead of either refractive or diffractive multifocality. The Crystalens HD was the first to fleetingly try this approach, though refractive predictability suffered. More recently, the Eyhance IOL has placed a stake in this ground. There may be promise in this category, yet dys­photopic complaints have shown up in this cohort of patients, albeit at a subjectively lower incidence.

Dyschromatopic IOLs. Most IOLs are colorless, though some have purposefully added so-called blue-blocking or violet-blocking chromophores to the IOL material, ostensibly to reduce risk to ret­inal health or reduce glare. Although some early cell-level bench studies suggest some benefits, the purported claims have still not been con­vincingly demonstrated, even now, two decades later. Related to this, some patients have had dyschromatopsias. One patient’s experience was well cataloged by Osher in the 2023 ASCRS Film Festival entry entitled “Yellow.” Cyanopsia has also been reported as an iatrogenic complication of the use of trypan blue in the setting of a hydrophilic acrylic IOL implantation, and the patient required an implant exchange.¹

EXPOSURE. In the photo at the left, it is not hard to imagine how this patient might see halos, arcs, shadow images, and glare due to the exposed aphakic space, IOL edges, and even capsulorrhexis margin, not to mention photophobia and contrast degradation from the excess light entering the posterior segment. On the right is the same eye after iris prosthesis placement in the capsular bag.

Capsule Dysphotopsias

Posterior capsular opacification (PCO) rates have been declining significantly over the last many years. Posterior capsule striae, however, remain a frequent—and often underdiagnosed—source of dysphotopsias. The most common presentation is for a patient to notice a streak with point sources of light, especially against a dark background. The patient rarely reports it in this manner and may just say “glare” or “starburst.” Identifying the axis of the stria on exam or with retinoscopy and que­rying the patient if the glare is along the orthog­onal axis (due to the Maddox rod effect) can be diagnostic. Some patients may actually see shadow images of letters when reading from a thick axial posterior capsule stria. Capsulotomy is therapeu­tic. Similarly, if a posterior capsulotomy is smaller than the scotopic resting pupil size, streaks can be seen from diffraction of the open edges, sometimes multiple. Enlarging the capsulotomy is the obvious solution.

Corneal Dysphotopsias

Many corneal irregularities can induce higher-order aberrations and/or irregular astigmatism. These may originate from epithelial layer base­ment membrane dystrophy or Salzmann nodules, curable by superficial keratectomy. Stromal issues like ectatic diseases, traumatic scars, or prior re­fractive surgeries have, depending on degree, been treated with scleral contact lenses, custom laser ablations, or, in some instances, full-thickness keratoplasties.

Gas permeable contact lens. Some higher-or­der aberrations originating from the cornea can be addressed by fitting a gas permeable contact lens.

Pinhole treatment. Over a half century ago, when iatrogenic irregular astigmatism from intra­capsular cataract surgeries ran rampant through the population, Peter Choyce, FRCS, developed a pinhole implant to treat these patients. He called it a “stenopeic” implant, and it was based on a rigid anterior chamber IOL platform.² When a wave of corneal decompensation and UGH (uveitis, glau­coma, hyphema) syndrome buried this platform, the genius of his stenopeic pinhole vanished.

More recently, pinhole optics regained new life, first with Morcher’s sulcus-based pinhole implants for keratoconus and other applications. Now Bausch & Lomb’s pinhole IOLs are available to the U.S. market, albeit they are labelled for presbyopia reduction rather than irregular astig­matism. Even suture-based pinhole pupilloplasty, popularized by Amar Agarwal, MS, FRCS, has been added to the anterior segment surgeon’s formidable war chest.³

LAL. While the Light-Adjustable Lens (LAL) has been approved for only spherocylindrical correction, it is interesting to wonder whether this technology or other customized IOL modifications might eventually be applicable for the optical correction (or compensation) of corneal higher-order aberrations.

Iris Dysphotopsias

There are numerous visual disturbances associated with iris defects that include haloes, shadows, crescents, and blurred or colored lines called “linear dysphotopsias.” Dysphotopsias from traumatic defects and even from laser iridoto­mies can induce streaks, glare, and even mon­ocular multiplopia, despite “20/20” vision. The Snellen trap strikes again! These patients fre­quently present after having seen other providers who have dismissed their complaints. With many surgical techniques for iris repair and ready access to iris prostheses—now even here in the United States—these patients’ vexing phenomena can be treated.

Opaque periphery contact lenses can provide temporary, partial amelioration of symptoms, though masks at the corneal plane can still permit stray light to access defects from non-axial light rays. Similarly, some providers utilize corneal tattoo; however, the inks are not labelled for this purpose, their color match can be challenging, and pigment migration over time may occur. Like opaque diaphragm contact lenses, corneal tattoos may induce restriction of peripheral vision in the meridians in which they act.

Key Points

Listening carefully to our patients’ concerns and keeping the different origins of pathologic dys­photopsias in our conscious “evoked set” are the crucial steps in sorting out how to best manage their dysphotopic complaints. We have many tools already in our armamentarium, and advancing technology will surely bring us more options.

___________________________

1 Werner L et al. J Cataract Refract Surg. 2002;28(7):1279-1286.

2 Choyce P. Intra-Ocular Lenses and Implants. H.K. Lewis & Co.;1964:24-26

3 Narang P et al. J Cataract Refract Surg. 2019;45(5):539-543.

ONCOLOGY AND PATHOLOGY

New Horizons for Precision Medicine

Patricia Chévez-Barrios, MD

The use of liquid biopsies and artificial intelligence (AI) in “multi-omics” will push personalized, pre­cision medicine forward in diagnosis, prognosis, and treatment of intraocular tumors.

Liquid Biopsies

A new approach to obtain a liquid biopsy of the eye’s aqueous humor in children with retinoblas­toma is paving the way for “cell-free” genomic sequencing of cancer tumors.¹ This is a welcome development because traditional needle biopsy shouldn’t be performed on a retinoblastoma due to the risk of spreading the tumor, which could lead to potentially fatal metastatic disease. An aqueous humor liquid biopsy that results in cell-free DNA (cfDNA) analysis and other biomarkers holds great promise in identifying the unique genetic profile of a given intraocular mass, leading to more precise diagnosis, personalized progno­sis, and tailored treatment. And peripheral blood liquid biopsy is also starting to be used in the early diagnosis and prognosis of metastatic disease in intraocular tumors.

Prognosis. Aqueous humor liquid biopsy—a biopsy without disturbing any tumor cells—can be evaluated for unique chromosomal changes that predict the probability of safely and success­fully saving a cancerous eye. Specifically, a gain of the chromosome called “6p” in the aqueous humor is associated with a 10-fold increased risk that an eye with retinoblastoma will require enucleation.¹ The ability to measure levels of the 6p chromosome holds great potential as a predictive biomarker to aid clinical decision-mak­ing when treating children with retinoblastomas. Similarly, other biomarkers in the aqueous humor may have prognostic value in various intraocular tumors.

Differential diagnosis. Aqueous humor biopsy will also assist in differential diagnoses. Tradi­tionally, ocular oncologists rely on ophthalmo­logic features and imaging to make a differential diagnosis of common intraocular tumors (e.g., retinoblastoma in children; and metastatic tumor, uveal melanoma, and lymphoma in adults) both malignant and benign. Since standard of care for patients with malignancies requires a biopsy-prov­en diagnosis prior to treatment, these new liquid biopsies will greatly enhance differential diagnoses in ocular oncology.

For example, a 2022 case study of a child with an atypical presentation of intraocular tumor (which was diagnosed as malignant ciliary body medulloepithelioma after enucleation) showed the promise of liquid biopsy. The aqueous humor biopsy taken from this eye demonstrated a different genomic signature not consistent with retinoblastoma, supporting the liquid biopsy’s value in the differential diagnosis of intraocular lesions when compared with traditional histology, MRI, B-scan ultrasound, and other diagnostic techniques.²

Treatment follow-up. These liquid biopsies also can be taken at multiple points to clearly show how tumor cells are responding to treatment. We can now detect pieces of DNA as well as smaller genetic markers of tumors floating in the aqueous humor, including messenger RNA, proteins, and extracellular vesicles, which are tiny pieces of cyto­plasm from the tumor. This allows us not only to interrogate a tumor at a single time point as with a needle biopsy but also to “tap” the aqueous humor in progressive liquid biopsies over time to see how the tumor is evolving and to monitor its response to therapy. The biomarkers in the aqueous humor correlate with the tumor itself.

After radiation plaque treatment for melano­ma, oncologists could potentially also evaluate treatment success by measuring DNA from the tumor in the aqueous humor; if it’s decreasing, the tumor is dying.

Given all these potential benefits, liquid biopsy deserves further clinical trials to validate its use.

BIOMARKERS IN FLUIDS. Cell-free DNA and other biomarkers found in the aqueous humor and/or blood make fluid biopsy a promising alternative to traditional diagnostic techniques in ocular oncology

“Multi-Omics” and AI

Precision medicine will also take huge leaps forward with the application of AI to the entire array of “-omics,” such as genomics, epigenomics, proteomics, and RNA transcriptomics. Taken together, these data are called “multi-omics,”³ and with advances in AI, researchers can now study multi-omics together to discover their correla­tions, associations—and implications for diagnosis and treatment.³

Vast -omics data. Cancers are now understood to be the result of a complex interplay among the multitude of -omics data in the human body, which contains an estimated 20,000 proteins, up to 22,000 protein-coding genes, 30,000 mRNAs, 114,100 metabolites, and other biological data.³ Immunolomics, also called immunomics, is an­other exciting type of -omics data⁴ from a cancer because it shows the environment the tumor is living in. This multi-omics data can be combined with data from any bodily fluid including blood, which carries the same bits of DNA, RNA, mRNA, and proteins that a tumor is producing. This knowledge advances precision medicine by treat­ing each patient’s unique type of tumor in each body’s unique environment.

With today’s digital slides and ample cloud storage, most pathologists now scan glass slides digitally on high-volume scanners, which is creating huge databases of digital slides. AI is beginning to combine these slide databases with multi-omics data, patient imaging, and clinical data to create specific groups of diseases with precise, effective therapeutics.

Precision medicine. While finding a single biomarker for a given cancer is exciting, real prog­ress in the future will likely come from finding patterns among multi-omics data to understand why one patient responds to treatment and a sim­ilar patient does not. That’s where AI is proving essential to the future of ocular oncology and pathology.

The type of AI called machine learning is now being applied to multi-omics data to classify cancers and their subtypes, to predict patient out­comes and prognosis, and even to identify novel therapeutic agents.³ Ultimately, even socioeconomic, geographic, and other personal data could be included in multi-omics research to fine-tune personalized medicine.

Specifically, a potential AI application now be­ing explored creates a barcode of a digital pathol­ogy slide and labels it as “cancer” or “not cancer,” and creates other barcodes for a patient’s MRI, CT scan, OCT, blood chemistries, and medical history. AI could then suggest a diagnosis, prog­nosis, and treatment for that patient, given his or her unique multi-omics data that are similar to others who responded well to a given treatment. That’s the point of personalized medicine: rather than treating everyone the same, to personalize care to individual patients.

Right now, multiple researchers are applying AI to multi-omics across several medical specialties.⁴ The goal is diagnosis and treatment integrated with everything known about a patient that culmi­nates in precision medicine. We are on the verge of this. In the next five years, we could see some of this coming to fruition.

___________________________

1 Berry JL et al. Mol Cancer Res. 2018:16(11):1701-1712.

2 Pike S et al. Ophthalmic Genet. 2022;43(6):855-861.

3 Biswas N, Chakrabarti S. Front Oncol. 2020;10:588221.

4 Rhee J et al. JACC CardioOncol. 2020;2(3):379-384.

TECHNOLOGY

The Predicted Future Impacts of ChatGPT

Given the explosion of artificial intelligence (AI) chatbot technology, EyeNet asked three ophthalmologists to predict the best uses of this technology for 2024 and beyond. Cecilia S. Lee, MD, MS, explains how generative AI chatbot technology like ChatGPT works. Thomas Hwang, MD, predicts its most useful—and safe—applications with patients. And Renee C. Bovelle, MD, considers its potential for assist­ing with time-consuming documentation tasks in the clinic.

How ChatGPT Works: The Promises—and Perils—of ChatBots

Cecilia S. Lee, MD

“GPT” stands for Generative Pretrained Trans­former, which is a type of large language model of AI that summarizes and assembles content. ChatGPT is a user interface that works on top of GPT. And it lets users pose questions, called “prompts,” and receive answers generated by GPT. As its name suggests, ChatGPT lets users interact with a given database of content in a conversational way, giving the impression of talking with another human.

The company OpenAI has created GPT and ChatGPT, with both nonprofit and profit divisions; Microsoft offers Bing; and Google’s Bard is another chatbot entering the market.

How GPT chatbots are built. Large language models are “trained” by inputting a large corpus of data and training the model to predict the next word by recognizing patterns in natural language. This iterative training process, for instance, might include showing the GPT model 10 “tokens,” which can be a word or part of a word, and then asking what the 11th token would be, based on all the text the model has already seen. It’s a pattern recognition process.

A misconception is that ChatGPT can “learn” if you correct a wrong answer it has made. It is not learning; users are simply providing more context to the model if they correct a wrong answer with follow-up prompts. The “learning” ended with the initial training of the AI model.

The latest version of GPT, called GPT-4, has been astounding. GPT-4 has been shown to act as if it has some human intelligence by seeming to reason through logical problems. A likely explanation here is that its training library included complex mathematical problems with their answers outlined, along with highly structured, logical computer code. After learning the patterns in these texts, GPT-4 is able to produce text that appears like human reasoning.

Hype and “hallucinations.” One of the main challenges for new users of any GPT model is understanding that it was only trained to mimic human language—and not trained to provide correct answers. There’s no true fact-checking ability to the model, which leads to the problem of so-called hallucinations, or content fabrications. For example, if you ask ChatGPT to provide references to support a medical fact it gave you, ChatGPT might give you a journal title with dates and page numbers—but that article does not actually exist. ChatGPT simply generated a sequence of “token” words or numbers that statistically, given its training, might come next.

The latest version of ChatGPT-4 [August 2023] hallucinates less than earlier versions, but the need for human beings to validate and fact-check the text it generates is still something that health care providers need to be aware of and researchers need to investigate.

So overall, we’re still in the hype phase with AI large language models in ophthalmology. And there are problems to be ironed out: along with developing mechanisms for content validation and correcting any false information that’s being relayed, another challenge is deciding who will be held liable when there’s an error. Currently, there are no standards for any kind of validation or agreement on liability for ChatGPT. As with past AI-driven screening tools for diabetic retinopa­thy, liability is an ongoing debate. A recent JAMA Health Forum editorial highlighted ChatGPT’s ability to mislead doctors, as well as its failure to source data transparently so that physicians can evaluate its generated texts’ validity.¹ There are further concerns about patients’ privacy and how much personal information can be input into such models.

Overall, in ophthalmology, there’s a lot of enthusiasm about adopting ChatGPT technolo­gy. But based on cautious adoption of previous machine-learning algorithms, it will take a few years to adopt—and we still must answer difficult questions about accuracy, liability, and privacy.

CARING. ChatGPT can be used in numerous capacities, but Dr. Hwang hopes that ophthalmologists will not choose to offload empathy to AI.

Predicted Uses With Patients: Trust, Errors, and Empathy

Thomas Hwang, MD

First, we as physicians must accept that chatbots like ChatGPT are a reality and that patients will use the technology to seek medical information, whether we like it or not. It’s similar to when patients started using the Google search engine, and providers were really worried about incorrect information. We’re still worried. But nobody’s thinking that patients won’t “google” medical con­ditions these days, and Google has made improve­ments to its algorithms since its launch.

Second, we as physicians need to understand the issues of trust and accuracy with ChatGPT technology. Most of the time, the technology is impressive, but then it will make fantastic errors and sound really good doing it. The algorithm doesn’t know if it’s making stuff up or not. For ex­ample, I was testing ChatGPT as a poten­tial way to tutor medical residents, and it gave me an incorrect answer to “the sum of 1.5 times 5.” My follow-up prompt was: “Doesn’t 1.5 times 5 equal 7.5?” ChatGPT answered, “No, that’s incorrect; 1.5 times 5 is 7.5.” So it was initially incorrect and then contradicted itself. We can’t yet fully trust the outputs of ChatGPT, whether as clinicians or as patients.

Patient communication. But I’m particularly excited about one ChatGPT application: its rewriting abilities for patient education materials. ChatGPT can rewrite patient instructions in dif­ferent reading levels and does a decent job of translation. Another future use for ChatGPT is in addressing simple patient questions that seem to pile up in physicians’ inboxes, a major cause of burnout these days. Today’s ChatGPT can answer only the most basic, factual questions, in part because it doesn’t ask follow-up questions as a human assistant would. But theoretically, an AI chatbot like ChatGPT could be developed to ease some of the patient communication burden.

Empathy research. JAMA Internal Medicine published a provocative paper comparing phy­sicians’ responses with ChatGPT responses to medical questions posted on a social media forum. A panel of health care providers ranked the two groups of responses based on “quality” and “em­pathy,” and the ChatGPT replies outperformed the physicians’ responses on both counts.² Personally, I have mixed feelings about that study because ChatGPT might sound more patient or polite, and write longer text responses, but that’s not real empathy.

To me, the ideal goal for AI is to give physi­cians more time and space to do uniquely human things like empathize, listen, and make high-level decisions that machines can’t be trained to make. I hope that we as physicians don’t take AI models such as ChatGPT as an opportunity to offload empathy, compassion, and communicating with our patients onto machines and technology.

Predicting Uses With Clinical Operations: Note-Taking and Drafting Appeals

Renee C. Bovelle, MD

Looking ahead at the pros and cons of ChatGPT, the clinical operations most affected will likely be writing patient notes and drafting letters seeking authorizations or appeals from insurers.

Patient notes. Some electronic health record systems are already incorporating ChatGPT into their note-writing capabilities, including the commonplace Epic. This use is still nascent, but it will progress very quickly. Some organizations use other technology, like Amazon Web Services, to transcribe patient visits, and then generate an automated patient summary using ChatGPT. The major players—Epic, CVS Health, and Carbon Health—are already refining these note-taking capabilities using ChatGPT.

Drafting letters. Sending out insurance letters for preauthorization or appeals is another area that will grow significantly in the near future. You can ask ChatGPT to generate a letter, for instance, stating why latanoprost isn’t the only medication to use for glaucoma. But to be clear, you still have to edit that draft and make sure everything is ac­curate because the technology scours the Internet, summarizes everything into readable form, and can make nearly anything sound true, even when it’s not.

Use care when uploading. Furthermore, to comply with HIPAA’s privacy requirements, patient information and protected health in­formation (PHI) should not be uploaded into ChatGPT or other AI systems. Similarly, copyright law precludes the upload of publications, such as research articles and textbook chapters, without first seeking permission from the owners.

How current is content? A challenge is that ChatGPT’s “training” database isn’t currently up to date [as of August 2023]. If you request the “lat­est” information, ChatGPT will give you material up to only September 2021. Eventually, it will be more current, but this is still new technology.

A work-around for this is to upload your own current content, taking care not to introduce any HIPAA-protected patient information or material that violates copyright law, or your own drafted bullet points from your website, and to ask ChatGPT to generate an insurance letter based on that. You then can edit the letter as needed and keep it as a template. You can even ask ChatGPT to write a letter based on that con­tent with an emphatic tone stating why a patient needs to be on a specific drug. If ChatGPT has generated a draft letter for you, be sure to review it yourself prior to submission to ensure its accuracy.

Marketing and education. ChatGPT also helps with social media marketing by drafting tweets and blurbs for Instagram. You can upload infor­mation on cataracts, for instance, and prompt it to generate 20 tweets based on that, which you then review and edit. You also can prompt ChatGPT to create marketing or patient education materials, and even have them written to a specific reading level or translated into Mandarin, Somali, Farsi, or other languages.

ChatGPT has great potential—but it’s not hu­man intelligence. Physicians are still needed.

___________________________

1 Mello MM, Guha N. JAMA Health Forum. 2023;4(5):e231938.

2 Ayers JW et al. JAMA Intern Med. 2023;183(6):589-596.

Meet the Experts

Renee C. Bovelle, MD Cornea Director at Howard University Ophthalmology Department in Washington, D.C., and in private practice in Glenn Dale, Md. Rele­vant financial disclosures: None.

Patricia Chévez-Barrios, MD Pro­fessor of Pathology and Labora­tory Medicine and Ophthalmology at the Institute for Academic Medicine at Houston Methodist and Weill Cornell Medical College in New York, N.Y. She is also Chair of Ocular Pathology and Program Director of the Ophthalmic Pathology Fellowship in the Department of Pathology and Genomic Medicine at Houston Methodist Hospi­tal. Relevant financial disclosures: None.

Thomas S. Hwang, MD A retina specialist and Kenneth C. Swan Endowed Professor of Ophthal­mology, Chief of the Retina Division, and Vice Chair for Education, Casey Eye Institute, Ore­gon Health and Science University in Portland. Relevant financial disclosures: None.

Cecilia S. Lee, MD, MS A medical retina specialist, Professor of Oph­thalmology, and Klorfine Family Endowed Chair at the University of Washington, Seattle. Relevant financial disclosures: None.

Michael E. Snyder, MD In private practice at Cincinnati Eye Institute and Professor of Ophthalmology at University of Cincinnati. Rele­vant financial disclosures: Alcon: I; Beyeonics Surgical: C; BVI Ophthalmics: C; DORC: C; EyeD Pharma: C; Haag-Streit: C; HumanOptics: C,P; Johnson & Johnson Vision: I; Trefoil Therapeutics: I; W.L. Gore: C.