As supply chain leaders, it is increasingly our responsibility to provide analyses to our stakeholders so they can make informed decisions which balance patient care and operational costs. Physician engagement, in particular, requires a significant amount of quality data regarding products, utilization, and their impact on both the organization’s profitability as well as on clinical outcomes. In this arena, our reputation is only as good as the last analysis we have presented.

We all know horror stories of how a major physician preference item (PPI) initiative fell apart after a key physician tore apart an analysis in front of an audience of her peers. Considering the data challenges we face in healthcare supply chain, this is an inevitable occurrence. Or is it?

Data stewardship is handling data in a responsible, consistent, and trustworthy way. Many data-intensive industries have recognized the need for employees whose role is to monitor and enforce good data practices. Even though this is not common within healthcare supply chain, there are many things we can do to ensure that we are good data stewards. A sample of key actions is described below.

The computer phrase "garbage in, garbage out" (GIGO) clearly states that sound data collection technique is the foundation of all our efforts. To ensure that this is a solid foundation, it is critical to know our systems. As supply chain leaders, we take it for granted that we inventory our products; yet, how many of us have ever inventoried the data that get collected and stored in our materials management information systems? In all our systems? A simple inventory of data fields, the type of data they contain and the data format can go a long way in ensuring that our analyses are consistently high quality.

In addition, it is important to check the data for critical errors. These are fundamental inaccuracies such as an anesthesiologist listed as the surgeon-of-record for a knee arthroplasty, an obstetrical procedure performed on a male patient or an implant log entry that records a physician implanting a brand they never used. After this quality check, it is also important to give the dataset the "smell" test to see if the data seem reasonable. If it does not, it will be necessary to review the completeness of the data source and/or eliminate outliers.

Once the data are collected, the analysis introduces new considerations. We must demonstrate that the final analysis accounts for all data we started with, including documentation of any outliers excluded from the analysis. This can be as simple as performing a count of records and a sum of totals before and after to ensure that we did not add or lose anything during our data transformation.

Even though most spreadsheet software allow us to condense many steps in one calculation, it is best to show our work and break chain calculations into their individual steps. This not only allows the audience of your analysis to follow the methodology, it also minimizes the risk of calculation errors. We should also adopt the practice of naming identical data with identical field names from one analysis to another and naming dissimilar data with different names. In this way, we are promoting transparency and consistency in our analyses both individually and as a whole.

Finally, once the analysis is complete, it is important to present it in an effective way. Our first responsibility is to know the audience, context and purpose of our presentation. What is most effective for an audience of administrators might be different than for physicians. In all cases, we are presenting to people whose time is a precious resource and often spend only a few minutes scanning the information; as a result, the presentation must make the purpose and result of our data immediately clear. This can often be accomplished by removing any unnecessary formatting and ensuring that any formatting used improves readability.

Graphs are also a common and powerful way to communicate information, but they are highly abused. We have all seen graphs that either confuse or misrepresent the underlying data. This can occur when unnecessary coloring, gradations and text obscure the meaning. It also occurs when graphs are not labeled or when the axes (the scale measuring the up-down and left-right parts of the graph) do not begin at zero.