Background: Abnormal reward processing is considered a core feature of severe mental illness (SMI). Deficits in this domain can have severe implications for daily functioning. The Online Positive Valence Study aimed to assess reward processing in bipolar disorder and schizophrenia using the National Institute of Mental Health (NIMH) Research Domain Criteria (RDoC) Positive Valence Systems (PVS) framework in an online task-based study. Recruitment and study adherence fell short of expectations. Objective: This paper critically evaluates the methodology used and shares the lessons learned from conducting an online task-based study of reward processing in a sample of patients with SMI. Methods: Using a cross-sectional design, the Online Positive Valence Study measured reward processing using 3 reward tasks, the monetary incentive delay task (MID), effort expenditure of rewards task (EEfRT), and a social and coin version of the Bandit task. Participants were eligible if they were 18 years or older. They were recruited mainly via email from 2 completed prospective cohorts. From 2023, the initial recruitment goal was no longer followed. Recruitment strategies and task performance are evaluated. Results: The sample consisted of 319 participants: 186 patients (58.3%), 81 first-degree relatives (25.4%), and 52 controls (16.3%). Multinomial logistic regression analyses revealed participants who were additionally called were more than 4 times more likely to respond “no” (odds ratio [OR] 4.57, 95% CI 3.12 to 6.70, <.001) and more than twice as likely to respond “yes” (OR 2.75, 95% CI 1.93 to 3.92, <.001) versus no response compared with those who received only emails. Personal salutations in invitation mails reduced the likelihood of participation compared to nonpersonal salutations (OR 0.31, 95% CI 0.20 to 0.47, <.001). Analyses of covariance (ANCOVAs) and linear mixed effects models (LLMs) revealed task performance was comparable between in-lab and online groups. Online participants showed longer reaction times on the MID task (=20.43, SE=2.55, =8.02, <.001) and lower win-stay rates on both Bandits (Bandit Coin: =–0.55, 95% CI –0.68 to –0.42, SE=0.06, =–8.73, <.001, η² =.14; Bandit Social: =–0.65, 95% CI –0.80 to –0.49, SE=0.08, =–8.54, <.001, η² =.14). Conclusions: For the first time, we performed a remote online task-based assessment of the PVS domain in patients with SMI, and we learned important lessons along the way. Future studies should carefully consider recruiting from existing cohorts; hire full-time dedicated personnel; and use personal approaches to recruitment, like phone calls, adjusting recruitment to the target population(s), and making participation technically as simple as possible. Attrition may be reduced by shortening the duration of participation and/or adding gamification. This study offers guidance for future research in online cognitive assessment in patients with SMI.
Background: Although bipolar disorder (BD) lacks objective biomarkers of illness progression, wearable devices enable continuous, real-world collection of physiological digital biosignals, offering a promising approach to address this gap. Measures such as electrodermal activity (EDA) and heart rate variability (HRV) fluctuate with mood state and symptom severity. Previous studies have reported reduced EDA during depressive episodes and increased HRV following symptom resolution after acute mood episodes. Furthermore, greater illness burden in BD has been associated with lower HRV, suggesting that digital biosignals may capture not only state-related changes but also aspects of illness course. Objective: This study investigated whether naturalistic EDA and heart rate (HR) profiles differ according to illness course variables in BD—including diagnosis, predominant polarity, years since onset, and lifetime number of episodes—in euthymic patients, to evaluate their potential as illness course biomarkers. Methods: This cross-sectional post hoc analysis used data from 80 euthymic patients on the BD spectrum from the INTREPIBD-TIMEBASE cohort. Participants were recruited at Hospital Clínic de Barcelona (Spain) between 2021 and 2024 and wore an Empatica E4 wristband for 48 hours in naturalistic settings. Inclusion criteria comprised a current or past (, Fifth Edition) diagnosis of BD confirmed by structured interview and sustained euthymia. We examined the effects of 4 illness course variables (diagnosis, predominant polarity, years since onset, and number of lifetime episodes) on 3 biosignals (mean EDA, EDA peaks per minute, and HR) using generalized additive models. Interactions between illness course variables and circadian patterns were also assessed. All models were adjusted for age, sex, and current antipsychotic use and included a participant-specific random effect. Results: Of the 80 participants, 49 (61.3%) were female; participants’ mean age was 46.9 (SD 14.1) years. After adjustment for covariates and multiple comparisons, no significant associations were observed between illness course variables and EDA or HR. In sensitivity analyses excluding patients with schizoaffective disorder (n=75), years since illness onset and predominant polarity were associated with EDA peaks per minute after multiple comparison correction. Specifically, patients with more than 22 years since illness onset had fewer EDA peaks compared with other groups (β=−1.06, 95% CI −1.71 to −0.41; =.001). No significant interactions were found between illness course variables and circadian patterns of EDA or HR. Conclusions: To our knowledge, this is the first study to examine the association between illness course variables and EDA and HR recorded in naturalistic settings in BD. In contrast to prior findings linking illness burden with HRV, our results suggest that years since illness onset may show modest associations with phasic EDA activity. These findings indicate that digital biosignals may be more sensitive to state-related changes than to illness course characteristics, limiting their use as longitudinal biomarkers of disease progression.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/99595c42a4915d3a2bbf77108a2dd7ca" />
Age-related macular degeneration (AMD) affects more than 5 million people around the world. And until now, no treatment could restore their vision. The Prima system, which was approved in Europe in July and is awaiting FDA clearance in the US, changes that. It combines a retinal implant with camera-equipped eyeglasses to improve sight in patients who develop a blind spot in the middle of their vision due to advanced AMD. After training sessions, more than 80% of recipients were able to read, do crosswords, and use their phones.
The retina processes light in three layers of cells: Photoreceptors convert light into electricity, bipolar cells process that signal, and ganglion cells transmit the signal to the optic nerve and the brain. Prima replaces the photoreceptors.
BROWN BIRD DESIGN
Developed by the California-based Science Corporation, the system takes advantage of the fact that AMD doesn’t damage all retinal circuitry; only photoreceptors—the cells that convert light into an electrical signal—die off. Prima’s implant mimics those cells and transmits visual information directly to healthy eye circuitry and on to the brain. “We’re leveraging all of the body’s own natural functions to process and transmit the information,” says Frank Brodie, an ophthalmologist at the University of California, San Francisco, and medical director for Prima.
BROWN BIRD DESIGN
CAPTURE
Glasses outfitted with a miniature front-facing camera capture what someone is seeing in real time. The video feed is roughly five megapixels running at 30 frames per second, which is similar to TV. The raw video is transmitted to a connected processor via a fiber-optic cable.
CONVERT
A small, handheld processor, which can be set on a table or tucked in a pocket, converts the video feed into infrared light signals that route back to the glasses. The device also includes physical buttons that let the wearer adjust contrast, brightness, and zoom.
BROWN BIRD DESIGN
DISPLAY
A laser projector on the glasses faces the wearer and shines pulses of near-infrared light directly at the eye at 880 nanometers—an invisible wavelength. This signal passes through the pupil and on to the retinal implant in the back of the eye.
SEE
The implant is made up of 378 pixels, each with an electrode and two photovoltaic cells, which do double duty. They convert the light into a current that healthy bipolar and ganglion cells pass on to the brain. They also act as solar cells to power the chip, eliminating the need for a battery.
Linda Marsa is a former Los Angeles Times reporter and an author who covers medicine, health, science, and the environment.
PHOENIX, AZ — In my fifth year attending the Cell and Gene Meeting on the Mesa, I’ve seen a handful of trends trudge forward, stop dead in their tracks, and, in some cases, reverse course.
Two years ago, cell and gene therapy (CGT) seemed like an untapped oil reserve waiting to power a new medical era. Casgevy, the first approved CRISPR medicine, had launched a new commercial era for gene editing, while CGT poster child CAR T-cell therapy was just one month shy of its seventh anniversary on the market, with seven approved therapies to match. All seven CAR T-cell therapies, along with Casgevy, were autologous.
An entire cell therapy industry formed around the autologous model, spawning and attracting a supersaturation of contract development and manufacturing organizations (CDMOs). To get a piece of the cell therapy commercialization pie, some CDMOs took up niches in preclinical development or clinical trial production at the requisite Current Good Manufacturing Practice (cGMP) standards. Other CDMOs specialized in the digital infrastructure needed to track cells on their “vein-to-vein” journey, and a select few even ventured into the realm of developing instruments for customers to bring production in-house.
Few cell manufacturing concepts attracted as much attention as Cellares, touting a new generation of CDMOs—an automated “Industry 4.0” CDMO model dubbed “intelligent development and manufacturing organization” (IDMO). Co-founder and CEO Fabian Gerlinghaus preached that the platform would be an autologous cell therapy “kingmaker.” That claim didn’t seem overconfident after Cellares bolstered its position in April 2024, signing a deal worth up to $380 million with Bristol Myers Squibb (BMS) to manufacture selected CAR T-cell therapies globally.
On the genetic medicine side, a few dozen gene therapies had been approved, and in vivo gene editing appeared imminent. Novartis’s Zolgensma, the standout, had generated roughly $1.2–1.4 billion annually since 2021 and had treated several thousand patients globally, while Sarepta’s Elevidys ramped much faster, from $200 million in 2023 to $821 million in 2024 and $899 million in 2025.
However, between 2024 and 2026, there have been major setbacks across CGT. Elevidys was poised to surpass Zolgensma, but safety concerns and label restrictions halted its growth in 2025 after three Sarepta gene therapy-induced acute liver failure deaths. In August 2026, BMS walked away from the manufacturing deal, claiming Cellares’s offering failed to meet “the necessary requirements to make commercial Breyanzi.”
Yet the tone at the 2026 Cell and Gene Meeting on the Mesa was anything but woeful or beaten down. Instead of personalizing every CGT, commercialization is pushing developers to broaden the patient pool for each program. These caught my eye at the Alliance for Regenerative Medicine (ARM) annual meeting.
Jodi A. Cook, PhD, CEO, Skylark Bio
Searching for the sweet spot
Jodi Cook, PhD, CEO of Skylark Bio, is a gene therapy veteran. She has been to all fifteen Cell and Gene Therapy Meeting on the Mesa conferences and told Inside Precision Medicine a pragmatic field is emerging. “The industry exploded on new money for a while,” Cook said. “Money was flowing everywhere, not just into gene therapy, and then it stopped. Now, we’re back to being mission-focused with sound business fundamentals to drive products forward.”
Instead of raising successive rounds around incremental milestones, Cook believes that companies need enough capital to reach the clinic and compelling data to unlock the next investment. “Now, it’s more like you raise one large round of funding to get into first-in-human trials,” Cook said. “That makes it increasingly important to establish the commercial opportunity much earlier.”
According to Cook, Skylark, a clinical-stage gene therapy company targeting inherited hearing loss and CNS disorders, prioritized commercialization and de-risking from the outset, advancing preclinical studies from mice to NHPs and then minipigs, whose inner ears closely approximate human size. “I tell my investors that we’ve de-risked everything we can externally,” said Cook. “Now the risk that’s left is execution, and I can control that.” Cook added, “All of the preclinical work and science line up. If you get it into the right cells, which we are, that should, over time, improve these children’s hearing, we did everything we could.”
Skylark’s lead program illustrates one way to balance precision with scale. It targets GJB2-associated hearing loss, which accounts for 15% to 30% of all inherited hearing impairment. But rather than developing treatments for individual mutations, Skylark delivers a functional copy of the entire GJB2 gene. “There are multiple different mutations within the GJB2 gene,” Cook said. “We’re replacing the whole gene, so we don’t have to slice and dice to personalize it to you.”
Skylark is applying the “doing more with less” approach throughout the pipeline. While each program requires a tailored payload, promoter, and regulatory elements, Cook is pushing for Skylark to get as close as possible to a “plug-and-play” model, reusing its capsid, manufacturing process, and much of its analytical infrastructure. That reuse could compress development timelines: what took three and a half years for Skylark’s first indication took about 18 months for its second.
The goal, Cook said, is to finally make good on a longstanding gene therapy promise: “The business model can work if you can reuse or reapply the platform instead of doing something completely different for every individual.”
One striking, out-of-the-box example of broadening a single gene therapy’s clinical applicability came from Nanoscope Therapeutics’ gene therapy solution to restore vision. Inherited retinal dystrophies (IRDs) are a diverse group of progressive genetic eye disorders characterized by the dysfunction and degeneration of photoreceptors (rods and cones) and the retinal pigment epithelium (RPE). Nanoscope’s strategy, however, has nothing to do with correcting the underlying congenital mutations that cause blindness. “If you don’t have functional photoreceptors, it may not matter which disease caused it, and there are dozens of them,” Samarendra Mohanty, PhD, Nanoscope’s president, CSO, and co-founder, said to Inside Precision Medicine.
Nanoscope’s MOGENRY (sonpiretigene isteparvovec) delivers an engineered chimeric multi-characteristic opsin (MCO) to surviving bipolar cells that critically transduce light-gated photoreceptor activity. COO Paul Hallen said, “We’re essentially hijacking the bipolar cells to make them de facto photoreceptors. If your photoreceptors are dying and your bipolar cells remain intact, and there are roughly 10 million of them, that creates an opportunity.”
Credit: ttsz/iStock/Getty Images Plus
To do so, Nanoscope doesn’t insert functional human opsins but rather a synthesized, Frankenstein-esque opsin that is far from naturally occurring or human. “There is no single molecule or opsin in nature that is broadband-sensitive, fast, and highly sensitive,” said Jared Stephens, Nanoscope’s vice president of strategy and business development.
The MCO combines components derived from two algae and a sea anemone to respond rapidly across ordinary ambient light, making it suitable for restoring real-world vision without blur. “Here, we’re expressing a protein that is not only non-mammalian but derived from three different non-mammalian organisms,” Mohanty added. “This may be the most synthetic protein or gene sequence ever expressed in a human.”
Nanoscope and other vision-loss companies call their tech optogenetics, which was intriguing after the 2026 Nobel Prize in Physiology or Medicine announcement. Mohanty mentioned Nanoscope’s optogenetic approach but, unprovoked, clarified the differences between their MCO and the work of newly minted Nobel laureates Karl Deisseroth, MD, PhD, Peter Hegemann, PhD, and Georg Nagel, PhD.
“In optogenetics research, if you’re trying to probe 100 different types of cells in the brain, the goal has primarily been to stimulate each with a single wavelength of light,” Mohanty said. “You want zero crosstalk: blue responds to blue, green to green, red to red, etc. Our goal was the opposite.”
Regardless, regulators are seeing the broad potential of Nanoscope’s MCOs. The PMDA—Japan’s FDA equivalent—recently accepted MOGENRY for priority review for IRDs, supported by data in retinitis pigmentosa and Stargardt disease.
Allan Reine, MD, CEO, Prime Medicine [Credit: Prime Medicine]
But what about gene editing?
For Allen Reine, MD, CEO of Prime Medicine, the key to CGT is to nail down differentiated products and platforms. “I don’t think we want to be another company simply doing another PCSK9 therapy,” said Reine. “Are we solving a problem that others can’t by doing that? Probably not. The question is, where is prime editing differentiated, and where is it a unique approach to solving a problem?”
The answer for Reine lies in addressing an unmet need with a differentiated therapy/approach that will make a meaningful difference to patients. “We have to check those boxes,” said Reine, who was an investor for 15 years before coming onto the operational side. “If all of those things are true, then there is likely a commercial opportunity. Then what we’re doing is good for patients and good for investors. When those two things align, that’s a great place for a company to be.”
Reine thinks their Wilson’s disease program qualifies, and on day 1 of the conference, Prime Medicine announced the first dose of PM577a, an investigational in vivo prime editor for H1069Q-mutated Wilson disease, in its global Phase I/II study.
Prime’s Alpha-1 Antitrypsin Deficiency (AATD) program trails close behind. Even though it’s a competitive space, Prime’s AATD program meets Reine’s criteria, who explained, “With alpha-1, it’s a competitive space, in part because the vast majority—95%, maybe 98%—of patients have one mutation. That makes it a very attractive target for gene-editing and RNA-editing therapies.”
Reine expects clinical results for the Wilson and AATD liver programs in 2027, which will prove useful for advancing programs targeting other cell types. Prime has shown promising preclinical data in various tissues and indications, including the lung and cystic fibrosis, which Reine believes is highly translatable and profitable. “I strongly believe we can ultimately show that there are good business models behind these therapies in many cases,” said Reine. “That’s where we’re going to see important inflection points. To me, that means more dollars flowing into the space, more dollars being spent on programs and drugs that ultimately get to patients, and more patients who can potentially be cured of disease.”
Fabian Gerlinghaus, co-founder and CEO, Cellares [Credit: Cellares]
Automated autologous cell therapies
Though disappointed in BMS’ withdrawal from Cellares’ manufacturing agreement, Gerlinghaus believes the company will stay on course. “Losing BMS was a setback, but BMS does not define the company,” Gerlinghaus told Inside Precision Medicine. Two weeks before the termination and reduction in force became public, he was visiting customers nationwide. Their response, he said, ranged from “We’re with you. This doesn’t change things for us” to asking whether Cellares could accelerate existing projects or take on additional trials. “One decision from one customer doesn’t change the vote of confidence we’ve gotten from a dozen other customers in the industry.”
Cellares has since leaned harder into a portfolio strategy, more than doubling its customer base since the beginning of the year and aiming to build a portfolio of 30 to 40 active autologous ex vivo programs. “I don’t see us slowing down. I actually see us accelerating,” Gerlinghaus said. The main consequence of losing BMS, he argued, is timing: “It’ll just take about a year longer to get to commercial manufacturing than it otherwise would have.”
There are successes for Cellares to report, including being one of eight companies picked for the FDA PreCheck program, which received over eighty applications. Similar to TSA PreCheck, the selection provides accelerated facility inspections for new domestic human drug and biological product manufacturing facilities ahead of commercial launch. Cellares was chosen alongside Eli Lilly, Regeneron, Amneal, Fujifilm Biotechnologies, Kriya Therapeutics, and Kyowa Kirin.
The setback also tempered some of Gerlinghaus’ swagger. Asked whether he still stood by his previous description of Cellares as a potential “kingmaker,” Gerlinghaus replied: “As I get older, I get more humble.” After years of seemingly uninterrupted milestones, he acknowledged, “Now we’ve had a setback. But show me one company that never had a setback.”
No company, field, product, or market has ever grown without setbacks. The key is to stay afloat until the stars align for a breakthrough.
In Part 2 of this two-part series, Tim Hunt, CEO of the Alliance for Regenerative Medicine (ARM), and Karim Mikhail, the new director at the Center for Biologics Evaluation and Research (CBER), explain existing and developing frameworks for growing a sustainable cell and gene therapy industry.
<strong>Background:</strong> Mental illness contributes substantially to global disability, and public adoption of AI for mental health support is accelerating without commensurate safety evaluation. General-purpose large language models can fail to recognize naturalistic text expressing suicidal ideation, with immediate clinical consequences. Domain-specific natural language processing methods offer a contrasting approach, but ensembles integrating the two have not been formally characterized. <strong>Objective:</strong> We aimed to (1) benchmark 5 modeling approaches for classifying mental health–related social media text, (2) develop and optimize ensembles integrating a domain-specific natural language processing classifier with a fine-tuned large language model, and (3) derive a closed-form framework constraining ensemble weighting in safety-critical classification. <strong>Methods:</strong> We analyzed 60,889 publicly available Reddit and Twitter texts spanning 9 categories (anxiety, bipolar disorder, depression, a normal baseline, personality disorder, stress, suicidal ideation, attention-deficit/hyperactivity disorder, and autism spectrum disorder). Labels were derived from the originating subreddit or a self-stated condition and are not clinical diagnoses. Five architectures were compared: prompt-engineered base GPT-4o-mini; fine-tuned GPT-4o-mini; a domain-specific classifier (support vector machine over symptom-informed lexical features); and 2 ensembles of these models, hybrid probability–indicator fusion and soft probability fusion. Data were split into 70%, 10%, and 20% at the post level (n=42,622, n=6089, and n=12,178); weights were selected on the validation split and all reported performance comes from the held-out test split. Closed-form bounds on permissible language model weights were derived for both strategies. CIs are paired bootstrap intervals (n=10,000 replicates); accuracy differences were tested with exact McNemar tests. <strong>Results:</strong> The domain-specific classifier achieved 91.9% accuracy (95% CI 91.4%-92.4%), exceeding the base (58.7%) and fine-tuned (88.6%) large language models. Both ensembles outperformed either constituent model, reaching 93.9% (soft fusion; 95% CI 93.5%-94.3%) and 93.8% (hybrid fusion; 95% CI 93.4%-94.2%) at a weighting of 60% classifier to 40% fine-tuned model, selected on validation (<i>P</i><.001 for both against the classifier alone; the 2 ensembles did not differ from each other [<i>P</i>=.16]). The hybrid ensemble reduced the miss rate for the suicidal ideation category from 32.5% (650/2000) to 1.9% (37/2000), a 17.6-fold reduction relative to the base model, although its advantage over the classifier alone within that category was not significant (<i>P</i>=.68). Accuracy fell discontinuously above a 50% language model weight, where the hybrid ensemble reduced exactly to the fine-tuned model; the selected weight satisfies the derived instance-level bound of 44.0%. <strong>Conclusions:</strong> Domain-specific clinical grounding remained necessary for stable ensemble performance in this corpus, and probabilistic fusion outperformed both constituent models when language model weight was bounded by a derivable, model-agnostic constraint. Because labels were community-inferred and only one corpus was analyzed, these results are hypotheses requiring external validation on clinically characterized data before use in decision support.
BackgroundSleep disturbances are common in severe mental illness, including schizophrenia, bipolar disorder, and major depressive disorder. Yet sex-specific sleep patterns across schizophrenia, bipolar disorder, and major depressive disorder remain poorly characterized.MethodsIn this cross-sectional study, we compared overnight polysomnography between 372 first-episode patients (aged 18-45) (93 schizophrenia, 181 major depressive disorder, 98 bipolar disorder). All participants underwent a single-night 10-channel polysomnography.ResultsAfter controlling for age, illness duration, BMI, and medications, women achieved significantly longer total sleep time (median 449.3 vs 420.3 min; p <0.001), time in bed (543.0 vs 525.5 min; p = 0.03), and higher sleep efficiency (83.3% vs 82.3%; p = 0.004) than men. Exploratory post-hoc analyses revealed potential sex-related differences in major depressive disorder and bipolar disorder (all p < 0.05, uncorrected for multiple comparisons). However, these findings did not survive correction for multiple comparisons, indicating that the observed differences should be interpreted with caution as preliminary.ConclusionsThis study found that biological sex modulated sleep macro-architecture in severe mental illness, with women exhibiting longer sleep duration and higher sleep efficiency relative to men. Although the cross-diagnosis comparison revealed only a marginally significant sex-by-diagnosis interaction that did not persist after multiple testing correction, these exploratory subgroup findings suggest potential nuanced variations in sleep architecture. Consequently, future studies with larger cohorts are warranted to evaluate whether sex-stratified approaches could refine clinical sleep interventions across psychiatric conditions.
Ketamine has progressed from solely being an anesthetic agent to a drug of rising clinical and research interest due to its distinctive pharmacological properties. Its dissociative properties and rapid-acting antidepressant effects have made it a prominent area of investigation across many medical and psychiatric applications. This systematic review examines ketamine’s history, mechanisms of action, clinical uses, and significant challenges. A systematic search of PubMed, SCOPUS, PsycINFO, and other sources, generated peer-reviewed publications that addressed the history of ketamine, its pharmacology, clinical applications, recreational use, safety considerations, and societal consequences. Overall, 106 articles were identified for inclusion in this study. At the receptor level, ketamine is an antagonist of NMDA receptors that triggers a number of downstream effects ranging from heightened glutamate transmission, AMPA receptor activation, increased BDNF expression, synaptogenesis, and mTOR pathway signaling. At the network level, it alters activity within large-scale brain circuits including the default mode network, affecting cognition, mood, and perception. Clinically, ketamine is used for anesthesia, treatment-resistant depression, and pain, with off-label uses for PTSD, anxiety disorders, and bipolar disorder. The review also examines current challenges related to ketamine’s recreational use, associated risks, sociocultural factors, and key clinical and ethical considerations, including safety, monitoring, and telehealth prescribing of oral and sublingual formulations. It integrates previously fragmented research from anesthesiology, neuroscience, pharmacology, psychiatry, and public health, consolidates clinical and ethical guidance, identifies critical gaps in current knowledge, and proposes directions for future research.
Background: Off-label drug prescribing is prevalent across medicine, including psychiatry, often due to unmet therapeutic needs and inadequate responses to standard treatments. The PSYHAMM (Psychotropes Hors Autorisation de Mise sur le Marché) project, funded by the French Research Agency, investigates these practices. To support this research, a clinical data warehouse (CDW) with advanced data analysis tools was developed and deployed. This system integrates both structured and unstructured data from electronic health records, facilitating comprehensive data analysis. The goal is to improve understanding, regulation, and safety of off-label drug use in psychiatry by providing insights into prescribing patterns and their impacts, ultimately contributing to better clinical guidelines and patient care. Objective: This study aimed to evaluate the precision (positive predictive value) of a CDW in identifying candidate off-label prescriptions in psychiatry among the cases automatically flagged by the system, rather than its overall accuracy, sensitivity, or specificity. Methods: The PSYHAMM data analysis involved a retrospective study of pathology-medication pairs to evaluate the precision of a computerized system among system-flagged cases. This system was compared with manual checks performed by a psychiatrist. The evaluation process included verifying if the condition identified by PSYHAMM was documented in the medical record, assessing diagnostic agreement with tolerance for schizoaffective disorders, and ensuring the identified treatment was current or prescribed in the past. Precision was measured as the number of relevant documents retrieved divided by the total number of documents proposed and was computed for the precise diagnosis, the broad diagnosis, and the identified treatment among the flagged cases. Results: The study analyzed 197 records, identifying 14 unique drug-pathology combinations. Bipolar disorder treated with sodium valproate represented the most cases (108/197, 54.8%), followed by schizophrenia treated with sodium valproate (37/197, 18.8%). The overall precision for detecting off-label situations was 51.3% (101/197). The precise diagnosis achieved a precision of 75.6% (149/197), while the broad diagnosis showed a higher precision of 84.8% (167/197). The identified treatment had a precision of 61.4% (121/197). The primary challenge was temporal discrepancies, such as distinguishing between acute and chronic conditions, which accounted for most of the 48.7% (96/197) of cases that were incorrectly classified. Conclusions: As a single-center, proof-of-concept evaluation, the PSYHAMM project demonstrates the potential of automated systems to support the identification of off-label prescriptions in psychiatry as a sensitive prescreening step requiring expert validation. The relatively high false-positive rate was driven mainly by temporal discrepancies (drugs prescribed before the index stay, discontinued during the stay, or only hypothetically mentioned) rather than by semantic errors. Future research should focus on integrating real-time data analytics and expanding to multiple institutions to improve the utility of off-label detection systems.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/8d87a9ea2c1630ed8abd7abffad0ae5f" />
