3 tips take fear out of data measurement
It’s statistics, not sadistics’
Home care agencies have a dilemma: They want to study their processes and outcomes, but have no model. "That is a good thing, because they start with an open field. But it is also a bad thing," says Emily Rhinehart, RN, MPH, CPHQ, CIC, director of quality management services at AIG Health Care Management Services, in Atlanta. The company, a subsidiary of AIG Insurance, consults with the health care industry on issues ranging from workers’ compensation to health care quality issues.
The lack of information causes fear for many agencies. Often, the only guidance they get is from physicians who demand they adhere to statistical standards derived from studies acceptable by academic journals. This, says Rhinehart, is unnecessary. "We are more interested in finding out trends so we can design effective programs."
She recommends home care agencies follow these three rules for effective and easy data measurement:
1. Know what you are studying.
Many agencies are bad at posing research questions, Rhinehart says. Some may note all the important process indicators, while others concentrate on outcomes. Both are equally important to the study. For example, with diabetics, the process indicators might concern the design of an educational program on using glucometers. The education in this example must be delivered within the three visits allowed by the health maintenance organization. Questions might ask what kind of program it is a computer-based program, one using only written material, or one containing a patient/caregiver discussion. Another question might be about whether the education was provided to the entire family, or just the patient, she says.
The outcomes indicators have to be broken down into physiologic, adverse, functional, and educational outcomes. The questions in those instances might relate respectively to blood sugar levels; admission to a hospital; ability to work at a daily job; and ability to express an understanding of the use of a glucometer and sliding scale of insulin.
2. Gather a good sample.
The problem with this part of the equation, says Rhinehart, is gathering a sample that is specific, but not too specific. "You have to focus on those diagnoses that are widespread in your agency, but not too widespread." Congestive heart failure, strokes, and diabetes are likely starting places, she says. "But don’t try to measure the activities of daily living for all of your Medicare patients. That’s too much."
Even with the main focuses, you have to pick carefully. "You don’t want to pick all diabetics, but maybe all newly diagnosed insulin-dependent diabetic patients over the age of 50," Rhinehart says.
3. Develop a good data collection tool.
"This is the biggest mistake people make," Rhinehart says. "You have to determine what demographic information you want, what outcomes you want to measure, and what are the variables." In most cases, the general information is the same the patient identifier (name or number), gender, age, and principal diagnosis. The other items should have a bearing on the outcomes, such as whether or not the patient speaks English, whether the person is sighted or not, whether the patient lives alone, lives with a spouse or other caregiver, or lives alone but has a caregiver.
There is a caveat, however, in some of the demographic questions. "You have to sit down and have agreement about what these things mean," she says. For example, English-speaking may mean one thing to one nurse and another to someone else. "You have to decide if it means the person is fluent, the person speaks some English, the person speaks none but understands, or the person is completely lacking in English skills."
The outcome portion is also deceptively easy, says Rhinehart. With a diabetic, it might be control of serum glucose. "But you have to define it very well. Are you talking about the first sugar test in the morning, the 4 p.m. sugar, or the 10 p.m. sugar? Are you studying it for a week? For a month? For two years?"
Variables might include what type of glucometer or other equipment the patient uses. "People can get bogged down in these and choose more variables than they need to," she warns. "But usually you only learn that lesson over time."
Once you have the tool, don’t just copy and distribute it. Rather, do a pilot test first and get feedback from your staff. "Only after you have done that should you roll it out."
Effective data management
While a study is only as good as the data involved, how you use the data is also important, says Rhinehart. "People can have all these great forms, but no one knows how to get the data from one form to another."
Before you start a study, make sure you have designed your data collection tools so they are of use to your end product. "Put them in your spreadsheet form before the final study but after the pilot," Rhinehart says. And don’t even think about doing this unless someone in your agency understands and can use computer spreadsheet programs.
She also recommends that the format you use include coding, rather than raw data. If your study is on the 7 a.m., 4 p.m., and 10 p.m. blood sugars over a 14-day period, you know there are 42 data elements on each patient. "Your form may have days one through 14 with three boxes for each day. Don’t just record the blood sugar, but code it."
"A" might mean that the blood sugar is within normal limits (90-130); "B" would be not within limits (greater than 130); and "C" could mean below normal limits (less than 90).
"If you just put in a raw number, you have to do more work on analysis later. You generally don’t care about the normal ones. This way, you can eliminate them from the study easily without having to decipher what each number means," Rhinehart says.
Once an initial sample is studied say, 30 patients in a particular area conduct follow-up studies with other same-sized groups of like patients at intervals. You can then compare the data based on patient demographic information, or even by caregiver. Rhinehart says this can not only point to positive trends, but also problems.
If four nurses each had 10 diabetic patients, and one of each group was out of control during the diabetes study, then you can make a determination that this is the normal range, says Rhinehart. "If you get a nurse who has patients out of control 30% of the time, then you know that the nurse may have a problem." Or if you find that younger patients are more likely to be out of control at their 10 p.m. sugar on Saturday nights, you might determine the social life of these patients is getting in the way of their care. You may decide they have to have more education. "Then you change the program, measure it again, and see if you improve," she says.
"Fancy data and statistical sophistication is not important," Rhinehart says. "The most important thing is good data collection based on good planning, good definitions, and a good data collection tool. Then you have a good study."
If you have a physician saying you have to determine the statistical validity of your sample, Rhinehart says you can tell that physician he or she is wrong with a sense of comfort. "We are interested in trends, and that just needs good data, not a pi square test."
(Editor’s note: Rhinehart has led a seminar on this topic twice for the Home and Health Association of Massachusetts in Boston. If you would like further information on sponsoring such a seminar at your organization, contact her at the number below.)
You have reached your article limit for the month. Subscribe now to access this article plus other member-only content.
- Award-winning Medical Content
- Latest Advances & Development in Medicine
- Unbiased Content