18.2 Research Design, Implementation and Evidence Interpretation

From Observation to a Question

A nurse observes a clinical problem, reviews what is known and formulates a question. The scientific method involves systematic observation, a testable question or hypothesis, a suitable design, data collection, analysis and interpretation. It also includes reporting methods and limitations so others can assess the result. A strong personal impression is a starting point, not evidence that an intervention works.

A focused question identifies the population, intervention or exposure, comparison and outcome. 'Does dialysis education work?' is too broad. 'Among new home-HD trainees, does a structured emergency simulation improve observed response competence compared with the current teaching process?' identifies a measurable issue. Define what 'competence' means before collecting results; changing the definition afterward can bias the conclusion.

Select the Design

Randomized trials assign interventions by a random process and can reduce selection bias. Blinding, when feasible, reduces expectation effects. A cohort study follows exposed and unexposed groups; a case-control study compares prior exposures among people with and without an outcome. Cross-sectional studies assess a point or period in time and usually cannot establish temporal causation. Qualitative studies explore experiences and meanings through systematic methods rather than attempting to estimate every effect numerically.

Choose the design for the question and ethical constraints. It would be unacceptable to withhold essential safety care merely to produce an untreated comparison group. A comparison can use accepted care or another ethical approach approved by the relevant review process.

Sampling and Measurement

Define inclusion criteria, recruitment and sample size planning. A convenience sample from one shift may not represent all patients. Patients who agree to participate may differ from those who decline. Attrition can distort results if those who leave have different outcomes. Report who was included and what information is missing.

Use reliable, valid measures. Reliability concerns consistency; validity concerns whether the measure represents the intended construct. Counting signed education forms may be reliable but does not validly measure emergency response competence. Standardize observation and consider observer disagreement. Clinical outcomes also need denominators and clear definitions: raw infection counts from units of different size are not directly comparable rates.

Implementation and Participant Protection

Develop the protocol before enrollment: procedures, outcomes, data handling, safety monitoring and stopping criteria. Obtain the institution's determination of research versus quality improvement and required ethics review. Investigators should not decide that review is unnecessary simply because the activity occurs in ordinary care. Informed consent, when required, explains purpose, risks, alternatives and voluntary participation in understandable language.

Protect privacy and avoid coercion. A patient's dialysis access or quality of treatment must not depend on agreeing to research. Staff who recruit their own patients should consider the power imbalance. Store data through approved systems with controlled access. Record deviations and adverse events rather than concealing them to preserve a favorable result.

Analysis and Interpretation

Statistical significance is not the same as clinical importance. A small measured difference may be statistically convincing but have little practical benefit. Confidence intervals indicate precision and should be considered with effect size and harms. Absolute risk difference often helps decision making more than a relative percentage alone. If infections decrease from four to two in one period, the relative reduction is 50%, but small counts and exposure differences make interpretation uncertain.

Association does not establish causation in an observational study. Confounding occurs when another factor relates to both exposure and outcome. For example, patients with catheters may differ in illness burden from those with fistulas. Adjustment can reduce some confounding but does not guarantee complete removal. Avoid turning one study into a universal treatment mandate.

Research and Quality Improvement

Quality improvement tests changes to local processes and often uses repeated cycles. Research seeks answers that may contribute to generalizable knowledge. The boundary is determined by purpose and methods through the institution's process, not simply by calling the project 'QAPI.' Both require responsible data handling and patient protection. A QI project can still need additional review.

Scenario

A unit introduces a new hub-care teaching program and infections fall. Before declaring causation, assess changes in catheter exposure, surveillance definitions, patient mix and other interventions. A controlled or carefully analyzed design may strengthen inference. The nurse reports what changed, the actual rates and uncertainty, then decides with the team whether further testing is needed. Honest interpretation makes evidence more useful than an exaggerated success claim.

Source: HHS Office for Human Research Protections.