Inconsistent time points at which measurements are taken

Correlation between measures taken at different time points become less meaningful

A measure of limb functionality may be obtained from POD1 thru POD7, but a measure of postoperative pain may be obtained only up to POD1.

Wrong measures obtained

Analyses are much less meaningful

Using the example directly above, it might be more meaningful not only to obtain postoperative pain through POD7 – but to obtain pain with a (specified) activity rather than pain at rest to correlate with functionality.

Asking patients to do calculations

Patients have their own postsurgical and other issues to worry about, hence the reliability and validity of their responses are limited

Just as patients would not be asked to calculate their own BMIs, they should not be asked to do any ‘mental calisthenics,’ e.g., “Compared to a week ago, how would you rate your current pain?”

Collecting data that are not in a ‘data dictionary’

This may confuse the statistician or data analyst who may then run inappropriate analyses

A data analyst may not know that ketorolac is only given IV at some hospitals (thus only to inpatients). Here, ketorolac would not be included with other medications for post-discharge pain management.

Missing values

Statistical tests and estimates of effect are less precise and may even be biased. Statisticians and data analysts are reticent to impute values when more than, say, 5-10% are missing.

Follow-up data may not be collected completely – either by the research assistants or by the subjects themselves. Thus it may be difficult or impossible to summarize total scores in an instrument. When it is known a priori that specific data are likely to be missing, statisticians can write in data imputation methods into the statistical analysis plan (SAP).

Continuous v categorical variables

Use of parametric v nonparametric analyses

The Modified Brief q5 asks “In the last 24 hours, how much relief have pain treatments of medications provided?” Responses are available for the patient to circle from 0% (no relief) to 100% (complete relief) in 10% increments. Strictly speaking, this is not a continuous variable (e.g., patients do not/can not respond with 69%, etc.), yet this variable is probably analyzed as continuous. A better idea is probably to categorize the variable as < 50% v ≥ 50% relief.

Parts v wholes

Analyses could very well be double-dipping (similar to multiple comparisons)

The OBAS q7 asks “How satisfied are you with your pain treatment during the past 24 h (0=not at all to 4=very much)?” Hence, it would be tempting to use this item to get at ‘satisfaction with anesthesia care.’ However, the OBAS total score is a summary of all 7 items derived by a particular algorithm, so using both OBAS q7 and the OBAS total score as separate analyses amounts to duplicity.