DNP 679 Module 6 Practice Paper Critique: Statistics in a Randomized Trial of Transitional Care Example

Twenty Percent Versus Thirty-Seven: A Practice Critique of the Statistics in a Randomized Trial of Nurse-Led Transitional Care

Student Name

Edson College of Nursing and Health Innovation, Arizona State University

DNP 679: Biostatistics: Principles of Statistical Inference

Instructor Name

Month Day, Year

Twenty Percent Versus Thirty-Seven: A Practice Critique of the Statistics in a Randomized Trial of Nurse-Led Transitional Care

The Study and Its Question

Naylor et al. (1999) asked whether a comprehensive discharge planning and home follow-up protocol delivered by advanced practice nurses would reduce readmissions and costs among hospitalized older adults judged likely to do poorly after going home. Patients aged 65 and older admitted to two urban academic hospitals with selected medical and surgical cardiac conditions were randomly assigned to the intervention or to usual discharge planning, and 363 were enrolled (177 intervention, 186 control). The primary outcome was readmission within 24 weeks of discharge, with secondary outcomes including time to first readmission, number of multiple readmissions, total hospital days and Medicare reimbursements.

Design and Sample

A randomized controlled design is the strongest choice for this question, because random assignment balances measured and unmeasured patient characteristics and supports a causal inference about the intervention (Polit & Beck, 2021). A reader should check the baseline table to confirm that the groups were similar on age, diagnosis and prior admissions, since chance imbalances can occur in a trial of this size. The sample was drawn from two academic hospitals in one city and limited to high-risk older adults, which helps the trial detect an effect but limits generalization to community hospitals, rural settings and lower-risk patients. A critic should also look for a stated power calculation; a sample of about 180 per group can detect a difference of the size observed for readmission, but it would have limited power for less frequent outcomes such as multiple readmissions.

Outcomes and Statistical Tests

Readmission within 24 weeks is a binary outcome, so comparing the proportion readmitted in each group with a chi-square test is appropriate. Time to first readmission is a time-to-event outcome, best analyzed with survival methods that handle patients who were never readmitted or were lost to follow-up. Hospital days and Medicare reimbursements are counts and costs, and such data are typically right-skewed, with most patients at or near zero and a few with long stays. Reporting group means for these variables, as the authors did for hospital days per patient (1.53 versus 4.09), can be pulled by a handful of long readmissions. A median with an interquartile range, or a test designed for skewed data, would help a reader judge how typical the difference was.

The Main Result, Recalculated

The authors reported that 20.3% of intervention patients and 37.1% of control patients were readmitted within 24 weeks (p < .001). A p value alone does not convey the size or precision of the effect, so the result can be expressed in more useful terms. The absolute risk reduction is 37.1% minus 20.3%, or 16.8 percentage points. Using the standard error of a difference between two proportions, the approximate 95% confidence interval runs from 7.7 to 25.9 percentage points. The relative risk is 0.20 / 0.37, about 0.55, meaning the intervention group's risk of readmission was roughly 45% lower. The number needed to treat is 1 / 0.168, about 6: for every six high-risk patients who received the protocol, one readmission was prevented over 24 weeks. Even at the cautious end of the interval, the benefit is substantial.

Follow-Up, Missing Data and Multiple Outcomes

Three further questions bear on the inference. First, older adults with cardiac conditions die or withdraw during a 24-week follow-up, and a critic should check whether every patient stayed in the arm assigned at randomization for the analysis, the intention-to-treat principle, and how deaths before readmission were handled, since a patient who dies cannot be readmitted. Second, the trial reports several outcomes, readmission, multiple readmissions, days, time to readmission and costs, so some significant results could arise by chance alone; naming a single primary outcome in advance, as this trial did with readmission, limits that risk. Third, the number needed to treat should carry its own interval, which can be calculated by inverting the limits of the risk difference: roughly 4 to 13 here (Altman, 1998).

Interpretation and Context

The authors' conclusion that the protocol reduced readmissions is well supported by the randomized design and the size of the effect. The cost finding, a halving of total Medicare reimbursements, is plausible given the reduction in hospital days, but cost totals are sensitive to a few expensive cases and would benefit from intervals. The trial is now more than 25 years old, and discharge practices, lengths of stay and Medicare payment rules have changed, so the size of the benefit in a current setting is uncertain. The trial's statistical case for nurse-led transitional care remains strong, and it explains why later models built on it.

References

Altman, D. G. (1998). Confidence intervals for the number needed to treat. BMJ, 317(7168), 1309-1312. https://doi.org/10.1136/bmj.317.7168.1309

Naylor, M. D., Brooten, D., Campbell, R., Jacobsen, B. S., Mezey, M. D., Pauly, M. V., & Schwartz, J. S. (1999). Comprehensive discharge planning and home follow-up of hospitalized elders: A randomized clinical trial. JAMA, 281(7), 613-620. https://doi.org/10.1001/jama.281.7.613

Polit, D. F., & Beck, C. T. (2021). Nursing research: Generating and assessing evidence for nursing practice (11th ed.). Wolters Kluwer.