Rows of dark bars on a printed grid, males to the left of center and females to the right.
Detail of a chart of deaths by age and sex in the United States in 1890, printed in 1898. Library of Congress.

Research

Norlowe Research works in two areas, healthcare and quantitative research.

Healthcare

Most of the firm’s healthcare questions concern patients in everyday practice, where treatment is less controlled than in a trial and fair comparisons are harder to make. Much depends on knowing who is counted and how each measure is defined.

Rounded red blood cells and several textured white cells, heavily magnified.A cylindrical ceramic jar painted with a leaping deer among leaves and flowers.
On the left, human blood cells under a scanning electron microscope, 1982, National Cancer Institute. On the right, a pharmacy jar made in Florence in the fifteenth century, The Metropolitan Museum of Art.

Chronic disease

How many adults have a long-term condition such as diabetes, how far the count depends on the definition used, and how it varies from one county to the next.

Data: National Health and Nutrition Examination Survey, PLACES local health estimates

Pharmacoepidemiology

How a medicine is used once it is approved, and what emerges when far more patients take it than any trial enrolled. Some effects are too rare or too delayed to appear any sooner.

Data: Medicare Part D Prescribers, FDA Adverse Event Reporting System

Trial evidence

The design and reporting of clinical trials, and the way the case for a treatment changes as further trials finish.

Data: ClinicalTrials.gov

Cost and coverage of care

Patterns in the use of health services and in what they cost, including how spending moves when coverage or prices change.

Data: Medical Expenditure Panel Survey

Quality measurement

How the measures used to compare hospitals are constructed, and how much of the difference they show between hospitals reflects the care given rather than differences in patients or chance.

Data: Hospital quality measures

Medicines in health data

How a medicine is named and coded on its label and in health data standards, and what is lost when one coding is translated into another.

Data: RxNorm, DailyMed

Quantitative research

Health data are usually collected to deliver care, to pay for it, or to meet a reporting rule, rather than to answer research questions. The quality of an estimate rests as much on how those data are cleaned and linked as on the model fitted to them.

A tall dark cabinet opened to show rows of counting wheels, with a printed sheet fixed inside its door.A city map crossed by a winding river, with darker shading spread over several districts.
On the left, a tabulating machine photographed in 1917, Library of Congress. On the right, a map of the cholera epidemic in Nashville, Tennessee, in 1873, National Library of Medicine.

Research design

How a question is turned into a comparison that existing data can answer. Who is eligible and when follow-up begins often matter more than the choice of model.

Case definitions

Which records identify a condition or an outcome, and how much an estimate changes when the definition does.

Data quality and linkage

Whether records from separate sources describe the same medicine or hospital, and how gaps in the data, such as counts withheld for privacy, can distort a total.

Statistical analysis

How a model is matched to the question and the data, and how the uncertainty in each estimate is carried into the conclusion. A ranking of hospitals, for example, can look more certain than the data behind it.

Methods notes

Data

The firm’s research draws on data released by United States government agencies. It has worked with the sources below.

Sources the firm has worked with
SourceKept byWhat it holdsUsed for
Medicare Part D PrescribersCenters for Medicare & Medicaid ServicesYearly totals of the prescriptions filled under Medicare Part D, listed by prescriber and by drug. Rows based on fewer than eleven claims are withheld.Pharmacoepidemiology
Hospital quality measuresCenters for Medicare & Medicaid ServicesMeasures reported for each hospital, among them death and readmission rates, safety events, and patients’ accounts of their stay.Quality measurement
National Health and Nutrition Examination SurveyNational Center for Health StatisticsA national survey that pairs household interviews with physical examinations and laboratory tests.Chronic disease
Medical Expenditure Panel SurveyAgency for Healthcare Research and QualityA national survey of households on the health services they use, what those services cost, and how they are paid for.Cost and coverage of care
FDA Adverse Event Reporting SystemFood and Drug AdministrationReports of suspected adverse events and medication errors, sent in by clinicians, patients, and manufacturers.Pharmacoepidemiology
ClinicalTrials.govNational Library of MedicineA registry of clinical trials, with the summary results that have been posted for completed trials.Trial evidence
PLACES local health estimatesCenters for Disease Control and PreventionModeled estimates of health measures for counties, cities and towns, census tracts, and ZIP Code Tabulation Areas.Chronic disease
RxNorm and DailyMedNational Library of MedicineRxNorm gives each clinical drug a standard name and identifier. DailyMed holds the labels that manufacturers file with the Food and Drug Administration.Medicines in health data

None of these sources identifies an individual patient. Each is used under the terms its agency sets.

Observational evidence

Many questions about treatment and care cannot be settled by a controlled experiment and have to be answered from observational evidence. Methods are chosen with that limitation in mind, and the assumptions behind each conclusion are stated. The firm also reports how far a conclusion moves when reasonable analytic choices are made differently.

Health data are shaped by the rules under which they are recorded, and those rules change. Definitions and measurements drift, and an effect that looks clear in one period or population can shrink or vanish in the next.

Many errors in health research come down to timing. In immortal time bias, one of the most common, patients are counted as treated from the start of follow-up because of a prescription filled later. They must survive until that prescription to be counted, so the treated group is credited with time in which it could not die.

Immortal time bias in research on medicines

Counted as treated from the start

  1. Follow-up begins
  2. First prescription
The arrow carries information from the later date back to the earlier one.
How each project is run