Table 4 Leave-one-site-out cross-validation site performance by model.

From: Individualized prediction of three- and six-year outcomes of psychosis in a longitudinal multicenter study: a machine learning approach

 

N

BAC Symptomatic outcome T31

N

BAC Symptomatic outcome T32

N

BAC Symptomatic outcome T61

N

BAC Symptomatic outcome T62

Amsterdam

81

60.8

104

69.5

81

69.3

104

60.1

Groningen

73

62.4

132

63.1

73

62.3

132

57.2

Maastricht

124

56.4

139

53.0

124

63.8

139

61.5

Utrecht

54

65.5

70

69.7

54

54.4

70

60.6

 

N

BAC Global outcome T31

N

BAC Global outcome T32

N

BAC Global outcome T61

N

BAC Global outcome T62

Amsterdam

80

66.4

100

68.1

77

63.9

98

65.4

Groningen

65

61.5

118

64.4

58

62.5

107

64.2

Maastricht

81

68.0

93

68.9

124

66.3

139

62.1

Utrecht

48

58.0

66

57.9

54

63.1

70

53.0

  1. Rows mention the geographic site left out of model training. Columns mention models organized per timepoint and included modalities. 1Models contained PANSS, demographic, illness-related, and CANSAS features; 2Models contained PANSS, demographic, illness-related, and CAPE features.
  2. BAC balanced accuracy, T3 follow-up at 3-year interval after the baseline, T6 is follow-up at 6-year interval after the baseline, PANSS Positive, and Negative Syndrome Scale, CANSAS Camberwell assessment of need short appraisal, CAPE community assessment of psychic experiences.
Back to article page