National trends in crowdfunding for neurosurgery: Scalable natural language processing of publicly available data...

Online crowdfunding, commonly used to cover healthcare costs for vulnerable populations, is directly linked to health disparities and gaps in social safety-net systems. The nationwide impact of crowdfunding on neurosurgery remains unclear. We aimed to characterize the funds raised, success rate, geographic distribution, and most frequent conditions for neurosurgery-related crowdfunding campaigns.

We obtained neurosurgical campaigns from the largest crowdfunding platform, GoFundMe (May 2010-December 2020). A large language model (GPT-4) was used to classify neurosurgical subspecialty and extract the main diagnosis. We used multivariable regression to investigate the association between the subspecialty and total funds raised, adjusting for social media characteristics and campaign characteristics (description length and campaign duration).

12,998 neurosurgical campaigns were included for final analysis. $514 million were sought for neurosurgical conditions, with $93 million raised (total funding percentage: 18.1 %). Crowdfunding campaigns requested on average $39,501.6 ± $1,244,929.7, raised $7,176.3 ± $12,380.5, and lasted 2.6 years ± 1.9 years (mean ± SD). In multivariable analysis, factors associated with increased funding include area of neurosurgical disease, campaign description length, and number of likes on social media. GPT-4 achieved an overall accuracy of 95 % on automated diagnosis extraction when compared to expert human reviewers on a 200-campaign sample with large (69x) speed increases.

We present the first analysis of crowdfunding for neurosurgery, an understudied and important funding source in the modern era. Funding success is significantly associated with a number of campaign-specific determinants. Moreover, we validate scalable large language models inference on complex textual medical information.