N-of-1 Trials in Nephrology

Nephrologists have long recognized the importance of conducting randomized clinical trials to assess treatment efficacy, but the translation of trial results into clinical practice can be challenging. Clinical trial participants may not be representative of “real-world” patients, given strict inclusion and exclusion criteria. Furthermore, within-group variation in effect size can be notable, impeding the accurate prediction of individual patient response. Given the economic toll and negative health effects stemming from inappropriate medication use, it is imperative to explore alternative strategies to augment treatment decisions.

In usual practice, treatment effects may be overestimated in the absence of predefined and objective criteria for treatment success. Reliance on trial-and-error or an optimistic case series may result in the unintentional use of a suboptimal treatment for years. N-of-1 trials present an opportunity to generate data and draw unbiased conclusions to inform personalized yet data-driven treatment decisions.

What Is an N-of-1 Trial?

N-of-1 trials are single-patient randomized trials that systematically compare treatments in a repeated crossover fashion. Protocols are designed to produce the least biased estimates, using randomization, blinded treatment allocation, objective outcome assessments, and balanced, repeated treatment periods. These features distinguish n-of-1 trials from the typical case report or case series and elevate this design to the top of the evidence-based medicine pyramid when appropriate design guidelines are followed (Figure 1).¹

Figure 1.

N-of-1 trial design protects against systematic and random error. This figure depicts a hypothetical scenario in which the treatment is ineffective but extraneous factors introduce systematic and random errors to outcome measurements, including lack of blinding, diet, stress, and the tendency of symptoms to wax and wane over time. In the top panel, a case report with pre-post results is shown. Unaware of confounders or placebo effects, the clinician incorrectly ascribes the change in outcomes to the treatment and mistakenly concludes that the treatment is effective. The bottom panel shows an identical disease course, but here, the treatment is administered in the context of an n-of-1 trial with repetition, randomized treatment order, and blinding. The multiple crossover design protects against systematic and random error, and the clinician correctly concludes that the treatment is ineffective.

Depending on the clinical question, multiple treatment options may be compared with each other, or a single treatment could be compared with placebo. The results are then used to inform treatment decisions for the individual patient but could also be aggregated across patients for generalizable knowledge.

Eligibility Criteria

Disease Characteristics

Not all conditions or treatment questions are amenable to an n-of-1 trial, but under the right circumstances, they can be a powerful tool to augment clinical decision making. The decision on whether to use an n-of-1 trial depends on the nature of the underlying condition, the treatment characteristics, and the outcome measures. The underlying condition should be chronic and require long-term treatment. Clinically relevant outcomes (evidence of treatment effect) should be easily measurable using objective criteria. Furthermore, the n-of-1 trial should be used only if the disease is expected to return to a baseline condition after each treatment is withdrawn. This approach is not appropriate if the treatments under question are expected to be curative or might permanently change the underlying condition.

For example, hypertension is an ideal use case for n-of-1 trials because hypertension is stable and long-term, antihypertensive medications have a relatively short onset and offset of action, and BP is easily measurable. N-of-1 trials have also been described in the setting of maintenance hemodialysis to compare strategies to reduce clotting events during dialysis.² Unsuitable clinical questions might include the comparison treatments with an unclear offset of action, such as dietary education interventions, or to test curative treatments for short-lived conditions, including the ideal choice of initial antibiotic for a dialysis catheter–related blood stream infection.

Treatment Characteristics

Any crossover design carries a risk for carryover effect, where the effects of a given treatment carry over into the next period, confounding the measured effects of the subsequent treatment. For these reasons, treatments with a relatively rapid onset and offset of action are the ideal candidates to be tested in an n-of-1 trial. A physical washout period during which no treatment is given will allow the previous treatment effects to dissipate and can reduce the risk of carryover effects. In circumstances when even a brief withholding of treatment would be unacceptable, an analytical washout period can be used. With this approach, there is no interruption between treatment periods and the analysis excludes any outcome measurements taken during the time when one might expect any overlap in the effects of consecutive treatments, based on the known pharmacokinetic profiles of the treatments.

Design Considerations

Several design parameters are important to consider. The number and length of treatment periods and the amount and frequency of outcome data collection are determined based on feasibility, acceptability to stakeholders, and the nature of the condition. Ultimately, these parameters will contribute to the accuracy of the conclusions and will determine how much experimentation is needed to produce a statistically valid treatment decision.

Data Collection and Analysis

Online platforms and mobile applications are being developed to facilitate remote data collection with built-in statistical computation.³ Interpretation of the data comparing treatment effects may be based on a simple graphical representation across treatment periods or a more formal statistical analysis using regression models that account for autocorrelation. If an n-of-1 trial protocol is conducted in a series of similar patients, inferences can be made on treatment effectiveness beyond the individual. Various meta-analytic methods have been described to aggregate data across participants,⁴ including individual participant data meta-analysis using multilevel models to account for repeated data from the individual nested within trials. A hierarchical linear model can provide an overall summary statistic across trials while allowing variability among trials. Bayesian models are a particularly appealing option that allow for straightforward interpretations of the probability of treatment outcomes at the individual or the population level.

Clinical Applications

Although n-of-1 trials were first proposed decades ago, their potential value is now more widely recognized, with notable examples in chronic pain management, lipid lowering therapy, and attention deficit hyperactivity disorder.^5–7 Nephrology-specific examples include informing the choice magnesium salts in patients with chronic renal magnesium wasting⁸ and in dialysate selection for a child with a rare metabolic disease.⁹

We previously reported the use n-of-1 trials in pediatric primary hypertension, testing three drugs and assessing which option produced the greatest reduction in ambulatory BP with the fewest side effects. The trials were feasible and acceptable, resulting in a change in treatment for 75% of participants. In a subsequent pilot randomized trial in adolescents, we observed a 93% probability that treatment informed by an n-of-1 trial produces greater BP reduction compared with usual care.¹⁰ This approach was less feasible in much younger children with secondary hypertension complicated by CKD, where we noted a lower proportion of caregivers and clinicians were willing to undergo a trial to modify hypertension management.

The potential for n-of-1 trials to optimize the management of nephrology patients remains promising. Therapeutic uncertainty plagues countless routine clinical decisions. The management of uremia-associated symptoms (e.g., pruritus or nausea), phosphate binder treatments, erythropoietin preparations, and dialysis prescription are all potential use cases for n-of-1 trials, which could directly improve patient-centered outcomes. One successful example tested three strategies to prevent dialysis circuit clotting when systemic anticoagulation was contraindicated, including intermittent saline flushes, continuous saline infusion, and heparin-coated dialyzers.² The three strategies were tested in a random order, and the time between dialysis sessions served as a washout period. Despite a very low sample size (n=46), investigators showed that heparin adsorption led to fewer dialysis interruptions than the other strategies.

Overcoming Barriers

Concerns about regulatory and ethical oversight might deter practicing clinicians from trying n-of-1 trials. Consideration of whether these could be classified as quality improvement or low-risk comparative effectiveness research could reduce regulatory burdens.¹¹ If answering common questions, protocols could be designed once and implemented repeatedly across similar patients within a network, such as a health care system or dialysis provider. This could directly improve the care of participating patients while contributing to generalizable knowledge for the broader nephrology population. Efforts are underway to develop an open-source digital tool, allowing protocols to be designed by clinical researchers and shared freely with clinicians through an online platform.³

Tackling uncertainty head-on with an n-of-1 trial could improve clinical outcomes and quality of life for many patients, without requiring the time and resources for a large-scale randomized trial to be designed, funded, completed, and published. It is time to move beyond case series and explore the judicious use of n-of-1 trials to improve the care of patients with kidney disease.

Disclosures

Disclosure forms, as provided by each author, are available with the online version of the article at http://links.lww.com/JSN/F464.

Author Contributions

Conceptualization: Joyce P. Samuel.

Funding acquisition: Joyce P. Samuel.

Writing – original draft: Joyce P. Samuel.

Writing – review & editing: Joyce P. Samuel.

Funding

This work was supported by Center for Clinical and Translational Sciences, University of Texas Health Science Center at Houston (1UM1TR004906-01).