Validation of key behaviourally based mental health diagnoses in administrative data: suicide attempt, alcohol abuse, illicit drug abuse and tobacco use

Table 2 Suicide attempt: comparison of administrative data using E-codes against suicide attempt notation in chart^c, both during the past 12 months prior to and including the index date

	*Unweighted Estimates, %* ^b *(N in fraction)*
*Samples*	*Agreement*	*Sensitivity*	*Specificity*	*PPV*	*NPV*
Suicide Death	88 (323/368)	17 (9/53)	100 (314/315)	90 (9/10)	88 (314/358)
Control	99 (359/362)	0 (0/3)	100 (359/359)	NA1	99 (359/362)
AD New User	97 (554/571)	11 (2/19)	100 (552/552)	100 (2/2)	97 (552/569)
Combined^a	93 (682/730)	16 (9/56)	100 (673/674)	90 (9/10)	94 (673/720)
	*Weighted Estimates, %* ^b *(95% confidence interval)*
Suicide Death	87 (82, 92)	19 (5, 32)	100 (100, 100)	96 (85, 100)	87 (82, 91)
Control	100 (99, 100)	0	100	NA1	100 (99, 1.00)
AD New User	98 (95, 100)	25 (0, 72)	100	NA2	98 (95, 100)
Combined^a	100 (99, 100)	2 (0, 5)	100 (100, 100)	NA1	100 (99, 100)

Abbreviation: NA1 is not appropriate (inappropriate to estimate because the N of the denominator is 0 for the unweighted estimate and thus the corresponding weighted estimates based partly or entirely on the control sample are also inappropriate); NA2 is not appropriate (inappropriate because weighting would be based on N = 2); AD is antidepressant; PPV is positive predictive value of an administrative diagnosis; NPV is the negative predictive value of an administrative diagnosis.
^a Combined sample of suicide deaths and controls
^b Rounded to a whole number
^c The chart notation had to be clear it was referring to a suicide attempt occurring in the past 12 months.

ISSN: 1472-6963