Anatomy of a Research Project: From Idea to Publication

Birth of an Idea

As a career researcher, I'm always looking for new areas to study. That's one of the main reasons I attend the annual American Diabetes Association meeting—to identify gaps in our knowledge of diabetes that I may be able to fill with new research.

In June 2016, I was at that meeting in New Orleans, and I was drawn to a session about the epidemiology of diabetic kidney disease (DKD).^[1] Dr Jonathan Shaw from the Baker Heart and Diabetes Institute in Melbourne, Australia, talked about "challenging the paradigm of DKD progression in diabetes."

We define DKD by low estimated glomerular filtration rate (eGFR) or the presence of albuminuria, he said, but these two measures are not the same. Each confers its own level of risk for nephropathy progression, and the combination of the two is particularly potent. Furthermore, each measure is a known risk factor for cardiovascular disease (CVD) and mortality, but little is known about how much they jointly and independently contribute to poor outcomes.

Thus, an idea was born. Dr Shaw's remarks identified a knowledge gap that I decided I would try to fill.

Obtaining Funding

My work is almost entirely funded with "soft money," meaning that I am expected to bring in money from outside to cover 100% of my salary and that of my research team. Sources include the National Institutes of Health (NIH), the Patient-Centered Outcomes Research Institute (PCORI), and contracts with the pharmaceutical industry.

Obtaining funding can be a considerable stumbling block. NIH and PCORI funding are highly competitive and can take 1-4 years to secure, if at all. Proposals presented to pharma must align with one of their products to have any appeal. In other words, no matter how wide the knowledge gap is that I identify, convincing those who hold the purse strings that it's worthy of funding is another matter altogether.

In this case, I was fortunate to have an existing contract with a pharmaceutical company to study the progression of eGFR-defined chronic kidney disease (CKD) in people with and without diabetes. Amending the contract to include additional analyses that combined eGFR with albuminuria and adding CVD and mortality outcomes seemed like an obvious approach. I was happy that they agreed.

A Straightforward Analysis, but...

We planned to identify patients with diabetes who had a serum creatinine value (to calculate eGFR) measured between 2006 and 2012 and a urine albumin-to-creatinine ratio (UACR) measured around the same time. We would use the test results to classify patients into six eGFR categories and three albuminuria categories specified in Kidney Disease: Improving Global Outcomes (KDIGO).^[2] Then we would follow patients in each of the resulting 18 categories until the end of 2016 to compare rates of hospitalization for CVD or death.

As straightforward as the plan was, the execution of it was not.

I conduct observational research using the electronic health records of members of a large integrated healthcare delivery system. Our data are rich and comprehensive, with reliable information from 1997 to essentially real time. And according to guidelines, serum creatinine and UACR should be measured at least annually in people with diabetes.^[3]

Because the health plan scores well on quality measures and serum creatinine tests are routinely ordered in our diabetes population, I expected to find plenty of patients to work with. However, the clinical definition of CKD requires a confirmatory eGFR when the initial value is < 60 mL/min/1.72 m², and we wanted to be as accurate as possible.

At an individual patient level, ordering a confirmatory test would be no problem, but searching for the tests in observational data by programming an algorithm that works in tens of thousands of people was no simple task. We ultimately identified about 32,000 diabetes patients with a qualifying and confirmed eGFR. Then it was simply a matter of finding a UACR measured at about the same time.

But these were real-world observational clinical data. Not all patients had a UACR, and some clinicians apparently assessed kidney damage with a urine protein dipstick rather than the UACR. Although previous studies had combined proteinuria and albuminuria data, we chose to keep our KDIGO albuminuria categories "pure."

This decision, coupled with the inherent incompleteness of observational data produced a final sample size of 16,678 patients, or slightly over half of what we expected. That may sound like plenty, but the cell sizes for some of the most severe categories were quite small.

What We Found

Mortality risk and hospitalization for CVD were greater for each higher eGFR regardless of level of albuminuria. They were likewise greater for each higher albuminuria category, regardless of level of eGFR. Risk was greatest when lower levels of eGFR co-occurred with higher levels of albuminuria.

Thus, each independently contributes to risk. But to accurately assess CVD and mortality risk, it is essential to include detailed eGFR and albuminuria values.

Wasn't this finding already well known? Not exactly. Although several studies have reported similar findings, including meta-analyses of over 1 million participants in general population and high-risk cohort studies,^[4,5,6,7] all of them left holes for us to fill.

Some previous studies combined UACR and dipstick proteinuria data, so the albuminuria categories were incongruent with the KDIGO categories.^[5,8,9] Two of the aforementioned meta-analyses reported results using eGFR categories that were not consistent with KDIGO,^[4,6] whereas data from the ADVANCE study did not include individuals with eGFR < 30 mL/min/1.73 m².^[10] Another meta-analysis evaluated eGFRs < 60 mL/min/1.73 m² without considering all eGFR categories individually.^[11]

In all of these studies, eGFR was either assessed from a single value or the study did not specify whether low eGFR values were confirmed with a second test at least 90 days later. Because use of a single eGFR overestimates CKD by more than twofold,^[12] a validating test is essential to avoid misclassification. Thus, we believed our study was unique and worthy of publication in a top-tier journal.

Getting Published

Believe me, we tried.

We wrote a cover letter describing why we believed our manuscript provided new, previously unknown results, and started with a high-impact general medicine journal, but we were rejected without review or explanation. We met the same fate with two diabetes specialty journals, a cardiovascular journal, and two nephrology journals.

Along the way, we speculated that despite our cover letter, the novelty of our work was not coming through, so we rewrote the abstract to better emphasize the novelty. We tried another diabetes journal, and although we were again rejected without review, the editor provided a rationale, commenting that the data were "basically confirmatory" of what is known.

In all, our paper was rejected by seven journals without review, presumably because the editors did not think the findings were novel. Finally, a diabetes specialty journal sent the manuscript out for review, and we still had to respond to "lack of novelty" critiques before the paper was finally accepted.

I'm happy to report that the study is now published,^[13] and we are grateful to the Journal of Diabetes and Its Complications for doing so.

Conclusion

From the birth of the idea to the acceptance of the manuscript, this was fairly typical of what has happened previously with my observational studies in diabetes epidemiology. Sometimes the analyses go more smoothly, and sometimes publication happens more readily. It can be simultaneously tedious and exciting.

On behalf of my coauthors, I hope readers recognize the importance and novelty of our findings and consider this a significant contribution to the literature.