Introduction
Materials and methods
Sample population
Survey content
Question | Response options | Response type |
---|---|---|
What is your primary subspecialty? | Spine; neurovascular; neuro-oncology; trauma; epilepsy, paediatric; peripheral nerve; neuro-intensive care; functional; other | Single choice; free text |
What setting do you primarily practice in? | Academic hospital; non-academic hospital; private practice; other | Single choice; free text |
What is your level of experience? | Resident; fellow; board-certified/attending; chairperson; medical student; other | Single choice; free text |
What is your gender? | Male; female | Single choice |
What age group are you in? | < 30 years; 30–40 years; 40–50 years; 50–60 years; > 60 years | Single choice |
What country are you currently based in? | List | Single choice |
In your clinical practice, have you ever made use of machine learning? | Yes, no | Single choice |
If yes: | ||
What have you used machine learning for? Please select any of the applicable | Shared decision-making/patient information; outcome prediction; prediction of complications: interpretation/quantification of imaging; grading of disease severity; diagnosis; other | Multi-choice; free text |
Please rate the importance of the following reasons for using machine learning from 1 to 4, based on your own clinical experience | ||
Improved preoperative surgical decision-making/treatment selection | 1 (Not important) to 4 (Highly important) | Single choice |
Improved anticipation of complications | 1 (Not important) to 4 (Highly important) | Single choice |
Objectivity in diagnosis/grading/risk assessment | 1 (Not important) to 4 (Highly important) | Single choice |
Improved shared decision-making/ patient information | 1 (Not important) to 4 (Highly important) | Single choice |
Time savings | 1 (Not important) to 4 (Highly important) | Single choice |
If no: | ||
Please rate the importance of the following reasons for not using machine learning from 1 to 4 | ||
Not personally convinced of added value | 1 (Not important) to 4 (Highly important) | Single choice |
Lack of skilled resources (staff, equipment) to develop a model | 1 (Not important) to 4 (Highly important) | Single choice |
Lack of data (quantity/quality) to develop a model | 1 (Not important) to 4 (Highly important) | Single choice |
Limited time to implement ML in clinical practice | 1 (Not important) to 4 (Highly important) | Single choice |
Limited affordability | 1 (Not important) to 4 (Highly important) | Single choice |
Difficulties in deciding which processes may benefit most from application of ML algorithms | 1 (Not important) to 4 (Highly important) | Single choice |
Lack of ML models for my indications | 1 (Not important) to 4 (Highly important) | Single choice |
In your research, have you ever made use of machine learning? | Yes; No; I do not engage in medical research | Single choice |
Statistical analysis
Results
Response rate and respondent characteristics
Characteristic | Value (n = 362) |
---|---|
Age groups, n (%) (years) | |
< 30 | 28 (7.7) |
30–40 | 118 (32.6) |
40–50 | 96 (26.5) |
50–60 | 61 (16.9) |
> 60 | 59 (16.3) |
Male gender, n (%) | 323 (89.2) |
Specialty, n (%) | |
Spine | 131 (36.2) |
Neuro-oncology | 64 (17.7) |
Neurovascular | 49 (13.5) |
Paediatric | 32 (8.8) |
Functional | 27 (7.5) |
Trauma | 16 (4.4) |
Epilepsy | 5 (1.4) |
Neuro-intensive care | 4 (1.1) |
Skull base | 1 (0.3) |
Peripheral nerve | 2 (0.6) |
Other | 31 (8.6) |
Work setting, n (%) | |
Academic hospital | 244 (67.4) |
Non-academic hospital | 56 (15.5) |
Private practice | 56 (15.5) |
Other | 6 (1.7) |
Level of experience, n (%) | |
Board-certified/attending | 217 (59.9) |
Resident | 69 (19.1) |
Chairperson | 41 (11.3) |
Fellow | 18 (5.0) |
Medical student | 8 (2.2) |
Other | 9 (2.5) |
Geographic origin, n (%) | |
North America | 250 (69.1) |
Europe | 68 (18.8) |
Asia and Pacific | 15 (4.1) |
Latin America | 18 (5.0) |
Middle East | 9 (2.5) |
Other | 2 (0.6) |
Use of machine learning in clinical practice, n (%) | 103 (28.5) |
Use of machine learning in research, n (%) | 108 (31.1) |
Machine learning in clinical practice and research
Domain | Region | p | ||||||
---|---|---|---|---|---|---|---|---|
Overall (n = 362) | North America (n = 250) | Europe (n = 68) | Latin America (n = 15) | Asia & Pacific (n = 18) | Middle East (n = 9) | Africa (n = 2) | ||
Clinical practice, n (%) | 103/362 (28.5) | 64 (25.6) | 21 (30.9) | 5 (33.3) | 8 (44.4) | 3 (33.3) | 2 (100.0) | 0.125 |
Clinical research, n (%)a | 108/347 (31.1) | 69/239 (28.9) | 27/67 (40.3) | 3/15 (20.0) | 6/16 (37.5) | 1/8 (12.5) | 2/2 (100.0) | 0.087 |
Application | Frequency, n (%) (n = 103) |
---|---|
Outcome prediction | 62 (60.2) |
Prediction of complications | 53 (51.5) |
Interpretation/quantification of imaging | 52 (50.5) |
Shared decision-making/patient information | 40 (38.8) |
Grading of disease severity | 39 (37.9) |
Diagnosis | 20 (19.4) |
Predictors of machine learning use
Variable | Clinical practice | Clinical research | ||||
---|---|---|---|---|---|---|
OR | 95% CI | p value | OR | 95% CI | p value | |
Age group | ||||||
< 30 | 1.21 | 0.52 to 2.74 | 0.658 | 1.33 | 0.55 to 3.19 | 0.520 |
30–40 | Reference | - | - | Reference | - | - |
40–50 | 0.97 | 0.41 to 2.2 | 0.938 | 1.33 | 0.56 to 3.17 | 0.520 |
50–60 | 1.62 | 0.71 to 3.7 | 0.248 | 0.85 | 0.33 to 2.1 | 0.730 |
> 60 | 1.82 | 0.47 to 6.93 | 0.382 | 3.25 | 0.78 to 13.7 | 0.110 |
Male gender | 0.97 | 0.43 to 2.27 | 0.935 | 2.19 | 0.89 to 5.94 | 0.100 |
Specialty | ||||||
Spine | Reference | - | - | Reference | - | - |
Neuro-oncology | 1.12 | 0.53 to 2.32 | 0.763 | 2.76 | 1.28 to 6.05 | 0.010* |
Neurovascular | 1.13 | 0.51 to 2.43 | 0.754 | 0.67 | 0.26 to 1.61 | 0.380 |
Paediatric | 0.58 | 0.19 to 1.57 | 0.301 | 1.00 | 0.33 to 2.85 | 0.997 |
Functional | 1.00 | 0.37 to 2.50 | 0.996 | 2.79 | 1.03 to 7.47 | 0.040* |
Trauma | 1.46 | 0.55 to 3.68 | 0.425 | 3.80 | 1.44 to 10.02 | 0.007* |
Epilepsy | 2.27 | 0.75 to 6.74 | 0.140 | 3.80 | 1.14 to 12.9 | 0.030* |
Neuro-intensive care | NA | NA | 0.991 | NA | NA | 0.990 |
Peripheral nerve | NA | NA | 0.993 | 2.82 | 0.11 to 75.5 | 0.570 |
Skull base | 1 | 0.05 to 8.93 | 0.997 | 2.01 | 0.09 to 20.12 | 0.480 |
Other | NA | NA | 0.995 | NA | NA | 0.990 |
Setting | ||||||
Academic hospital | Reference | - | - | Reference | - | - |
Non-academic hospital | 0.67 | 0.30 to 1.43 | 0.315 | 0.23 | 0.08 to 0.57 | 0.003* |
Private practice | 0.59 | 0.26 to 1.28 | 0.195 | 0.36 | 0.14 to 0.85 | 0.026* |
Other | 1.11 | 0.13 to 6.89 | 0.915 | NA | NA | 0.990 |
Experience | ||||||
Board-certified/attending | Reference | - | - | Reference | - | - |
Resident | 1.40 | 0.56 to 3.6 | 0.458 | 1.14 | 0.44 to 3.00 | 0.790 |
Chairperson | 1.58 | 0.68 to 3.58 | 0.279 | 2.03 | 0.80 to 5.17 | 0.130 |
Fellow | 1.36 | 0.38 to 4.63 | 0.628 | 0.42 | 0.08 to 1.79 | 0.270 |
Medical student | 1.18 | 0.17 to 7.37 | 0.860 | 1.10 | 0.17 to 8.04 | 0.920 |
Other | 0.77 | 0.11 to 3.69 | 0.767 | 1.60 | 0.27 to 8.07 | 0.570 |
Geographic origin | ||||||
North America | Reference | - | - | Reference | - | - |
Europe | 1.12 | 0.57 to 2.16 | 0.738 | 1.32 | 0.65 to 2.63 | 0.440 |
Latin America | 2.48 | 0.81 to 7.52 | 0.547 | 0.49 | 0.10 to 1.83 | 0.330 |
Asia and Pacific | 1.43 | 0.41 to 4.46 | 0.106 | 1.42 | 0.35 | 0.630 |
Middle East | 1.64 | 0.30 to 7.45 | 0.536 | 0.16 | 0.01 to 1.15 | 0.110 |
Other | NA | NA | 0.992 | NA | NA | 0.999 |
Attitudes towards machine learning in neurosurgery
Region | ||||||||
---|---|---|---|---|---|---|---|---|
All | North America | Europe | Asia and Pacific | Latin America | Middle East | Africa | p value | |
Reasons for use | ||||||||
Improved preoperative surgical decision-making/treatment selection | 3.27 ± 0.86 | 3.14 ± 0.92 | 3.57 ± 0.6 | 3.6 ± 0.55 | 3.5 ± 0.76 | 3 ± 1.41 | 3 ± 1.41 | 0.430 |
Improved anticipation of complications | 3.13 ± 0.92 | 2.92 ± 0.96 | 3.57 ± 0.6 | 3.2 ± 0.84 | 3.62 ± 0.74 | 3 ± 1.41 | 3 ± 1.41 | 0.048* |
Objectivity in diagnosis/grading/risk assessment | 3.22 ± 0.84 | 3.25 ± 0.85 | 3.05 ± 0.74 | 3.4 ± 0.55 | 3.5 ± 0.76 | 3 ± 1.41 | 2.15 ± 2.12 | 0.680 |
Improved shared decision-making/patient information | 3.07 ± 0.9 | 3.06 ± 0.97 | 3.14 ± 0.65 | 2.8 ± 0.84 | 3.38 ± 0.74 | 2.5 ± 0.71 | 2.5 ± 2.12 | 0.720 |
Time savings | 2.62 ± 1.07 | 2.72 ± 1.03 | 2.29 ± 1.1 | 2.8 ± 1.1 | 2.5 ± 1.2 | 3 ± 1.41 | 2.5 ± 2.12 | 0.720 |
Reasons for non-use | ||||||||
Not personally convinced of added value | 2.04 ± 1.05 | 2.13 ± 1.05 | 1.77 ± 1.07 | 2 ± 0.94 | 1.56 ± 0.73 | 2.5 ± 1.22 | NA | 0.070 |
Lack of skilled resources (staff, equipment) to develop a model | 3.11 ± 0.98 | 3.14 ± 0.97 | 3.02 ± 1.07 | 3.1 ± 1.1 | 2.78 ± 0.83 | 3.33 ± 0.82 | NA | 0.670 |
Lack of data (quantity/quality) to develop a model | 2.67 ± 0.99 | 2.67 ± 0.99 | 2.72 ± 0.99 | 2.8 ± 0.92 | 1.78 ± 0.67 | 3.33 ± 0.82 | NA | 0.160 |
Limited time to implement ML in clinical practice | 2.85 ± 0.96 | 2.85 ± 0.98 | 2.98 ± 0.94 | 2.9 ± 0.88 | 2.33 ± 0.71 | 2.33 ± 0.52 | NA | 0.160 |
Limited affordability | 2.74 ± 1.08 | 2.77 ± 1.06 | 2.51 ± 1.16 | 2.5 ± 0.85 | 3.22 ± 1.09 | 3.33 ± 1.03 | NA | 0.034* |
Difficulties in deciding which processes may benefit most from the application of ML algorithms | 2.75 ± 0.96 | 2.77 ± 0.93 | 2.64 ± 1.11 | 2.6 ± 0.97 | 2.78 ± 0.83 | 3 ± 0.89 | NA | 0.900 |
Lack of ML models for my indications | 2.84 ± 1 | 2.82 ± 0.99 | 2.79 ± 1.12 | 2.7 ± 0.67 | 3.44 ± 0.73 | 3.33 ± 0.82 | NA | 0.250 |