TABLE 3.

Best performing individual representations based on group and finding level AUC

Finding	Prop.	P-Value	N-Grams	Document LIRE	Rules	Document MIMIC	Controlled Vocabulary	Controlled Vocabulary Filter Only
All Findings	-	1.06E-24†	0.960 (0.949, 0.972)✦	0.910 (0.892, 0.929)✧	0.897 (0.882, 0.911)	0.894 (0.872, 0.916)	0.882 (0.862, 0.901)	0.879 (0.857, 0.902)
Potentially Clinically Important Findings	-	2.06E-13†	0.954 (0.925, 0.983)✦	0.910 (0.878, 0.942)✧	0.821 (0.789, 0.852)	0.888 (0.856, 0.920)	0.857 (0.821, 0.894)	0.854 (0.813, 0.895)
any degeneration	0.896	0.03289	0.947 (0.906, 0.989)✦	0.896 (0.820, 0.972)	0.850 (0.770, 0.931)	0.874 (0.789, 0.958)	0.936 (0.911, 0.961)✧	0.913 (0.882, 0.943)
facet degeneration	0.762	0.13484	0.970 (0.940, 0.999)✦	0.963 (0.935, 0.991)✧	0.873 (0.832, 0.914)	0.949 (0.919, 0.979)	0.923 (0.888, 0.959)	0.922 (0.883, 0.961)
disc height loss	0.507	5.46E-10†	0.931 (0.891, 0.970)✦	0.833 (0.791, 0.875)	0.877 (0.829, 0.925)	0.830 (0.774, 0.886)	0.874 (0.812, 0.935)	0.878 (0.826, 0.930)✧
any stenosis*	0.480	0.09604	0.972 (0.950, 0.994)✦	0.957 (0.930, 0.983)	0.893 (0.856, 0.930)	0.967 (0.945, 0.988)✧	0.955 (0.926, 0.984)	0.961 (0.935, 0.987)
disc bulge	0.435	0.00276	0.986 (0.972, 1.000)✦	0.978 (0.956, 1.000)✧	0.976 (0.953, 1.000)	0.967 (0.939, 0.994)	0.955 (0.923, 0.987)	0.952 (0.920, 0.984)
foraminal stenosis*	0.400	0.00153†	0.950 (0.922, 0.978)✦	0.935 (0.899, 0.971)	0.914 (0.876, 0.952)	0.936 (0.906, 0.965)✧	0.932 (0.898, 0.966)	0.930 (0.896, 0.964)
central stenosis*	0.351	6.37E-10†	0.950 (0.919, 0.981)✦	0.903 (0.876, 0.930)	0.773 (0.731, 0.815)	0.915 (0.884, 0.947)✧	0.907 (0.861, 0.953)	0.901 (0.852, 0.950)
any osteophyte	0.332	3.44E-07†	0.955 (0.916, 0.994)✦	0.888 (0.845, 0.932)	0.925 (0.894, 0.955)✧	0.886 (0.835, 0.937)	0.875 (0.818, 0.933)	0.880 (0.819, 0.94)
listhesis grade 1	0.324	0.03888	0.967 (0.941, 0.994)✦	0.927 (0.880, 0.974)	0.958 (0.928, 0.989)✧	0.910 (0.858, 0.961)	0.945 (0.903, 0.987)	0.947 (0.904, 0.989)
disc degeneration	0.322	0.00203	0.935 (0.898, 0.973)✦	0.830 (0.792, 0.869)	0.904 (0.865, 0.944)	0.784 (0.726, 0.842)	0.909 (0.842, 0.976)✧	0.909 (0.863, 0.954)✧
scoliosis	0.274	0.03726	0.969 (0.943, 0.994)✦	0.924 (0.888, 0.961)	0.959 (0.924, 0.994)✧	0.929 (0.887, 0.971)	0.900 (0.850, 0.949)	0.901 (0.854, 0.948)
osteophyte anterior column	0.271	5.13E-10†	0.953 (0.930, 0.976)✦	0.867 (0.820, 0.914)	0.913 (0.872, 0.954)✧	0.846 (0.785, 0.907)	0.882 (0.831, 0.933)	0.874 (0.825, 0.923)
spondylosis	0.217	0.39974	0.992 (0.977, 1.010)✦	0.936 (0.881, 0.99)	0.990 (0.974, 1.010)✧	0.901 (0.846, 0.955)	0.900 (0.836, 0.964)	0.883 (0.817, 0.949)
fracture	0.212	0.62647	0.949 (0.912, 0.987)✦	0.947 (0.921, 0.973)✧	0.883 (0.816, 0.95)	0.896 (0.839, 0.953)	0.914 (0.868, 0.960)	0.925 (0.878, 0.972)
disc protrusion	0.197	0.69348	0.977 (0.953, 1.000)	0.940 (0.906, 0.973)	0.927 (0.879, 0.974)	0.910 (0.866, 0.954)	0.982 (0.952, 1.010)✦	0.980 (0.948, 1.010)✧
disc desiccation	0.189	1.39E-05†	0.981 (0.953, 1.010)✦	0.957 (0.923, 0.990)	0.958 (0.921, 0.994)✧	0.923 (0.884, 0.962)	0.817 (0.745, 0.888)	0.822 (0.747, 0.898)
nerve root displaced/compressed*	0.169	1.08E-06†	0.955 (0.913, 0.996)✦	0.913 (0.854, 0.972)✧	0.785 (0.698, 0.872)	0.907 (0.848, 0.966)	0.870 (0.819, 0.921)	0.864 (0.817, 0.911)
lateral recess stenosis*	0.163	0.00235	0.966 (0.921, 1.010)✦	0.941 (0.909, 0.973)	0.649 (0.567, 0.731)	0.948 (0.918, 0.978)✧	0.843 (0.771, 0.914)	0.844 (0.773, 0.915)
annular fissure	0.099	0.38358	0.950 (0.888, 1.010)✧	0.944 (0.886, 1.000)	0.957 (0.905, 1.010)✦	0.922 (0.848, 0.996)	0.763 (0.667, 0.860)	0.755 (0.650, 0.860)
nerve root contact	0.097	1.13E-08†	0.972 (0.949, 0.996)✦	0.921 (0.859, 0.982)✧	0.910 (0.827, 0.993)	0.914 (0.856, 0.973)	0.797 (0.716, 0.877)	0.818 (0.741, 0.894)
disc extrusion*	0.079	0.00089†	0.994 (0.963, 1.020)✦	0.979 (0.954, 1.000)✧	0.886 (0.775, 0.997)	0.948 (0.901, 0.995)	0.969 (0.914, 1.020)	0.962 (0.897, 1.030)
endplate edema*	0.059	0.00088†	0.916 (0.831, 1.000)✦	0.868 (0.765, 0.970)✧	0.854 (0.732, 0.976)	0.838 (0.713, 0.963)	0.789 (0.683, 0.896)	0.778 (0.626, 0.930)
hemangioma	0.049	0.0736	0.991 (0.950, 1.030)✧	0.899 (0.807, 0.991)	0.999 (0.995, 1.000)✦	0.875 (0.705, 1.050)	0.963 (0.867, 1.060)	0.951 (0.845, 1.060)
disc herniation	0.038	0.04629	0.956 (0.878, 1.003)✧	0.890 (0.715, 1.060)	0.975 (0.927, 1.020)✦	0.912 (0.740, 1.080)	0.760 (0.526, 0.994)	0.786 (0.520, 1.050)
spondylolysis	0.032	0.00567	0.981 (0.940, 1.020)✦	0.832 (0.654, 1.010)	0.936 (0.795, 1.080)✧	0.866 (0.747, 0.986)	0.754 (0.507, 1.000)	0.785 (0.561, 1.010)
listhesis grade 2*	0.028	0.00018†	0.905 (0.741, 1.070)✦	0.799 (0.604, 0.994)✧	0.767 (0.580, 0.955)	0.683 (0.486, 0.88)	0.735 (0.559, 0.91)	0.706 (0.510, 0.902)

For each representation, we trained and tested 26 models (one for each finding) on 80% and 20% of the dataset, respectively. For group level, we averaged the AUC across all findings and across all potentially clinically important findings for each representation. We repeated this process 25 times with different splits of the data to calculate 95% confidence intervals. Table shows the best performing representation ordered left to right based on the All Findings row (1^st row). The first column indicates the finding/group. For the findings, the second column indicates prevalence in the test set represented as a proportion. For each row, ✦ indicates best performing individual representation based on the average AUC and ^✧ indicates the second-best representation. We show the 95% confidence interval in the parentheses. Finally, for each finding and group, we performed a t-test comparing the best representation’s distribution of AUC values for the 25 repeats to the second-best representation. † indicates significant comparisons with Bonferroni correction (p-value significance for the groups: 0.025 = 0.05/2 groups, p-value significance for the findings: 0.0019 = 0.05/26 findings). Finally * indicates the potentially clinically important findings. AUC = Area Under the Curve, Prop. = Proportion.