Table 5.

Threats To Internal Validity (Adapted From Robson, Shannon, Goldenhar, And Hale (2001), TABLE 3.1).

Threat to internal validity	Description	Likely impact for Intervention 1.2	Likely impact for Intervention 3.2
Dropout	Overall characteristics of intervention group change due to some participants dropping out, possibly affecting outcome	Several patients died during the study period and their data was excluded from the relevant datasets. This is not expected to affect the overall result. There were no other dropouts.	Several patients died during the study period and their data was excluded from the relevant datasets. This is not expected to affect the overall result. There were no other dropouts.
Hawthorne	Involvement of outsiders could affect outcome independent of the main intervention component (eg, because participants know they are being observed)	Pilot phase: Staff on the trial wards during the pilot could have tried especially hard to avoid discharge delays because they knew there was a trial underway, although some of the staff involved in the discharge process will have been unaware of the trial. Rollout phase: It is unlikely that the staff involved felt they were being measured.	If the Hawthorne effect significantly influenced the length of stay results, we would expect to see this most strongly near the start of the study, as sensitivity to being observed is likely to fall over time (Robson et al., 2001). In fact, we find the reverse – that average length of stay is lower in the second half of the period in which the SPRING form is introduced than in the first.
History	Other events may take place during the trial and influence the results	There were no known initiatives (other than those described in this paper) during the hospital at the time that could be expected to have directly impacted length of stay. Seasonal variations and secular trends in length of stay were explored in the control wards and found to be negligible as discussed below	There were no known initiatives (other than those described in this paper) during the hospital at the time that could be expected to have directly impacted length of stay. Seasonal variations and secular trends in length of stay were explored in the control wards and found to be negligible as discussed below.
Instrumentation	Measurement method or its validity changes during the intervention	Measurement method and its validity were constant throughout the study.	Measurement method and its validity were constant throughout the study.
Maturation	Intervention group develops in ways independent of the intervention, possibly affecting the outcome	This was not relevant in this study.	This was not relevant in this study.
Placebo	Participants believe that an intervention has material efficacy, even where is none	This was not relevant in this study.	This was not relevant in this study.
Regression to the mean (RTM)	Basis for choosing the intervention group is a greater need for the intervention; this would be expected to naturally change towards a normal value	Pilot phase: There is potential for RTM, since the two wards selected for the pilot were known to have been experiencing particular problems processing HNAs in a timely manner. Rollout phase: Since the intervention was implemented across the whole hospital, RTM should not be relevant.	The three wards chosen to implement the SPRING form showed fairly typical performance before the trial, with mean length of stay of 23.1 days (SD = 16.0); for all eight wards, the mean was 23.3 days (SD = 15.8). We therefore expect the RTM effect to be negligible in this case.
Testing	Taking measurements could have an effect on the outcome	This was not relevant in this study.	This was not relevant in this study.