Historical and future trends in emergency pituitary referrals: a machine learning analysis

Area of Science:

• Neurosurgery
• Endocrinology
• Data Science

Background:

• Acute pituitary referrals to neurosurgical services are common and require emergency care.
• Current patterns of pituitary emergency referrals and their prospective changes are not well-characterized.
• Understanding referral trends is crucial for effective service planning.

Purpose of the Study:

• To evaluate historical and current pituitary referral patterns.
• To utilize machine learning for predicting future service utilization.
• To identify key drivers and trends in acute pituitary referrals.

Main Methods:

• Analysis of electronic neurosurgical referrals (2014-2021) to a major UK pituitary center.
• Characterization of referral patterns, including common presentations and lesion types.
• Prediction of referral volumes using STL-ARIMA, compared against CNN-LSTM, Prophet, and baseline models.

Main Results:

• 462 emergency pituitary referrals analyzed; common presentations include headache (47%) and visual field deficits (32%).
• STL-ARIMA model demonstrated superior accuracy in predicting yearly referral volumes compared to other algorithms.
• Referral volumes showed a significant increasing trend and were unaffected by the COVID-19 pandemic.

Conclusions:

• This study is the first to use large-scale data and machine learning to analyze and predict acute pituitary referral volumes.
• Findings provide insights into future service demands and the impact of system stressors like the COVID-19 pandemic.
• The results highlight opportunities for service improvement in managing acute pituitary conditions.