Table 1 Clinical, demographic, and visit characteristic predictor variables for the regression

Patient also had self-harm diagnosis at time of index episode (0 = no, 1 = yes)

Did the patient have an additional ICD-10 code of X60 to X84, but excluding X65

.110***

Patient also had assault diagnosis at time of index episode (0 = no, 1 = yes)

Did the patient have an additional ICD-10 code of X85 to X99, Y00 to Y09, or T741?

.017

[21]

Patient also had “undetermined intent” diagnosis at time of index episode (0 = no, 1 = yes)

Did the patient have an additional ICD-10 code of X60 to X84, or Y10 to Y34? If the patient had an ambiguous coding – i.e., an “undetermined intent” code mixed with a self-harm or an assault diagnosis, then it was coded as being undetermined intent only. Also if the patient had a mixed assault and self-harm diagnosis, they were also coded as undetermined intent only.

-.002

Patient also had injury diagnosis at time of index episode (0 = no, 1 = yes)

Did the patient have an additional ICD-10 code of S*, T00-T32, T66, T67, T70, T71, T75, V*, W*, X00-X39, X50-X52? Excluded from this variable were cases where the patient also had another diagnosis which suggested that the injury was not a straightforward accident – i.e., a parallel coding of self-harm; assault; or event of undetermined intent.

-.042***

[21]

Patient also had other mental health diagnosis at time of index episode (0 = no, 1 = yes)

Did the patient have an additional ICD-10 code of F0 to F9, but excluding F7?

.061***

[7, 10, 16, 17]

Patient also had other substance use diagnosis at time of index episode (0 = no, 1 = yes)

Did the patient have an additional ICD-10 code of F11 to F19, T40, or X42?

.053***

[7, 10, 16, 17]

Patient’s primary diagnosis at index admission was alcohol related (0 = no, 1 = yes)

Was the first (primary) diagnosis recorded in HES for this admission an alcohol related code, or was this recorded as a secondary code?

-.087***

[20]

Average level of wealth of the patient’s area of residence at time of index admission (interval variable, ranging from 1 to 10)

Calculated from Index of Multiple Deprivation (IMD) coding of the patient’s address, and taken from the demographic fields of the HES index admission record

-.062***

[22]

Age at start of index admission (interval variable, ranging from 12-18 years)

Recorded on the index admission record as a demographic field

.032**

Patient was treated during index admission in an urban Primary Care Trust (PCT) (0 = no, 1 = yes)

Based on a pre-existing methodology and codings [31] and the classifications of the Office of National Statistics, did the patient have their index admission in a hospital that was in an “urban” PCT area, as opposed to a PCT in a “rural” area?

.009

Patient gender (0 = female, 1 = male)

Recorded on the index admission record as a demographic field

-.047***

[7, 10]

Address missing at time of index admission (0 = no, 1 = yes)

Was there a missing address at the time of the patient’s current admission – as indicated by a missing Local Authority code listed against their record?

-.003

Length of stay in hospital during the index admission (continuous variable, expressed in days)

Admission date subtracted from discharge data, expressed in days

.030**

Patient had index admission on a “traditional” drinking day (0 = no, 1 = yes)

Did the patient have their index admission on a day that was: (a) a Friday, Saturday, or Sunday, (b) a public holiday, or (c) a day on the eve of a public holiday?

-.043***

[23, 24]

Outside ambient air temperature on date of index admission (continuous variable, expressed in degrees Celsius)

The mean daytime temperature on the date of admission according to the Hadley Centre Central England Temperature (HadCET) dataset [32]

.008