Predicting therapeutic neurosurgical intervention within 7 days after initial nonoperative management of mild traumatic brain injury with subdural hematoma: development and internal validation of a clinical-radiographic model

When a person suffers a blow to the head that leaves them conscious and alert, doctors face a difficult balancing act. The injury might be minor, requiring only rest and observation, or it could be a ticking time bomb, where a small collection of blood inside the skull slowly expands and threatens the brain. This is the reality for patients with mild traumatic brain injury who also have a subdural hematoma, a specific type of bleeding that forms between the brain and its protective covering. While many of these patients can be safely managed without immediate surgery, a significant minority will deteriorate, requiring an operation to remove the blood and relieve pressure. The challenge for clinicians is knowing which patients fall into the safe group and which ones need closer, more urgent monitoring. Without a clear way to predict who will get worse, hospitals often resort to repeating scans for everyone or watching everyone closely, a process that is resource-intensive and can cause unnecessary anxiety.

Researchers at Linkou Chang Gung Memorial Hospital in Taiwan set out to solve this uncertainty by building a tool to predict which patients would need surgery within the first week of their hospital stay. They looked back at the records of 745 adults who arrived at their emergency department with a mild head injury and a subdural hematoma but were stable enough to be treated without immediate surgery. The team wanted to create a model that could take the information available right when a patient walks in—their age, their level of consciousness, and the details of their brain scan—and calculate the likelihood that they would require a surgical operation within seven days. The goal was not to decide who should have surgery, but to help doctors decide who needs the most intense surveillance and the earliest follow-up checks.

The study focused on a specific group of patients: those who were conscious enough to answer questions and follow commands, with a Glasgow Coma Scale score between 13 and 15, and who had been cleared for non-surgical observation by a neurosurgeon. The researchers excluded anyone who needed emergency surgery the moment they arrived. They tracked these 745 individuals to see who ended up undergoing a therapeutic procedure, such as drilling a hole in the skull to drain blood or removing a section of the skull to relieve pressure, within seven days of their arrival. In total, 111 of these patients, or about 15 percent, required such an intervention. The remaining 85 percent remained stable enough to avoid surgery during that first week.

To build their prediction tool, the team examined a wide range of factors that were present at the time of admission. They considered the patient's age, whether they had high blood pressure or diabetes, and if they were taking blood-thinning medications like aspirin. Crucially, they analyzed the initial brain scan in detail, measuring the thickness of the blood clot, checking for other types of bleeding in the brain, and noting the presence of any bruising or damage to the brain tissue itself. They used a statistical method designed to handle small numbers of rare events to ensure their calculations remained accurate even for uncommon findings.

The results revealed a clear hierarchy of risk factors. The single most powerful predictor of whether a patient would need surgery was the thickness of the subdural hematoma. For every 5-millimeter increase in the thickness of the blood clot, the odds of requiring surgery within a week increased dramatically. This finding held true even when the researchers accounted for other variables. Beyond the clot size, the presence of other injuries on the scan also mattered. Patients who had an epidural hematoma, a collection of blood between the skull and the brain, or those with bruising and bleeding within the brain tissue itself, faced higher risks. Similarly, the presence of blood in the ventricles, the fluid-filled spaces deep inside the brain, was a strong warning sign. Interestingly, patients who were fully alert with a perfect score of 15 on the consciousness scale were less likely to need surgery than those with a slightly lower score of 13 or 14, though even those with perfect scores were not entirely free of risk.

The researchers combined these factors into a single model that could estimate a patient's risk. When they tested this model, it proved quite effective at distinguishing between those who would need surgery and those who would not. The model successfully sorted patients into different risk groups. Those with the lowest predicted risk had a chance of needing surgery of only about 3 percent. In contrast, patients in the highest risk category faced a chance of 62.5 percent. This wide gap suggests the tool can effectively separate the low-risk patients, who might be safely observed with standard care, from the high-risk patients who need immediate, intensive monitoring and perhaps earlier repeat scans.

However, the authors were careful to define the limits of what their model could do. They emphasized that the model predicts the likelihood of a treatment decision made by surgeons at their specific hospital, not the inevitable biological progression of the injury itself. In other words, the model reflects the local practice of when surgeons decide to operate, which can vary from one hospital to another based on different thresholds for intervention. It does not tell a doctor that a patient's brain is definitely getting worse, only that under the current local practices, this patient is likely to be taken to the operating room soon. The model is a guide for surveillance, not a substitute for a surgeon's judgment.

The study also looked at patients who were missed by the model. A small number of patients who were predicted to be low risk still ended up needing surgery. In these cases, the patients often started with very thin blood clots and were fully alert, but they later developed new symptoms or showed changes on follow-up scans that were not visible at the moment of admission. Conversely, some patients predicted to be at very high risk did not need surgery within the week, often because they remained stable despite having large clots or because their surgery was delayed beyond the seven-day window. These exceptions highlight that while the model is a powerful tool for the initial assessment, the condition of a patient with a head injury is dynamic and requires continuous re-evaluation.

Ultimately, this research provides a structured way to think about risk for a common and potentially dangerous injury. By quantifying the influence of clot thickness and other specific scan findings, the model offers a data-driven way to prioritize resources. It suggests that doctors can be more confident in watching low-risk patients closely without immediate intervention, while directing their most vigilant attention and earliest follow-up imaging toward those with the highest calculated risk. The authors conclude that while this tool is a significant step forward for their hospital, it must be tested in other medical centers with different practices before it can be used widely. The next step is to see if the model holds up when applied to different populations, ensuring that it remains a reliable compass for navigating the uncertainty of mild traumatic brain injury.