NIHSS Stroke Database is Incomplete and May Have Selection Bias

Mathew J. Reeves BVSc, PhD, FAHA Professor, Department of Epidemiology and Biostatistics, Michigan State University East Lansing, MI 48824

Prof. Reeves Interview with:
Mathew J. Reeves BVSc, PhD, FAHA
Professor, Department of Epidemiology and Biostatistics,
Michigan State University
East Lansing, MI 48824 

Medical Research: What is the background for this study?

Dr. Reeves: The National Institutes of Health Stroke Scale (NIHSS) is the single most important prognostic factor in predicting outcomes of individual stroke patients. NIHSS data is obviously important at the patient level but also at a hospital level since the case mix of stroke patients are assumed to vary widely across different hospitals and referral centers.

Measuring stroke outcomes at a hospital level is becoming increasingly important as work proceeds in the US to develop integrated stroke systems of care. But it is also very relevant to the new payment models being introduced by CMS which are based on hospital rankings that are developed from statistical risk adjustment models. One would expect that NIHSS would be a major contributor to these models but currently a major limitation is that NIHSS is incompletely documented in clinical registries such as GWTG-Stroke, and is completely absent from administrative data.

The problem of missing NIHSS data plays havoc with the ability to risk adjust stroke outcomes across hospitals. Missing data results is a smaller number of stroke cases being included in the risk adjusted calculations for a given hospital which results in greater uncertainty over what the actual hospital outcomes are. Further there is concern that NIHSS data is not missing at random, and so the NIHSS data that is documented may represent a biased selection of all the cases that a hospital admits. This too could have important consequences for hospital rankings.

To determine the degree of potential bias in the documentation of NIHSS data this study examined trends in and predictors of documentation of NIHSS across 10 years of data (2003-2012) in the GWTG-Stroke program.

Medical Research: What are the main findings?

Dr. Reeves:

  • The overall NIHSS documentation rate was 56% but between 2003 and 2012 documentation increased from 27% to 70%. NIHSS scores were higher at hospitals with lower NIHSS documentation rates – a pattern that suggests that NIHSS data were subject to modest selection bias.
  • NIHSS documentation rates were higher in patients who arrived within 3 hours of symptom onset and in those who arrived by ambulance indicating that NIHSS was more likely to be recorded in patients who were eligible for tPA treatment.
  • As documentation improved in more recent years the degree of selection bias in NIHSS scores lessened indicating that the problem of selection bias is improving as NIHSS documentation rates improve.

Medical Research: What should clinicians and patients take away from your report?

Dr. Reeves: It is important that hospitals record the NIHSS score on arrival for all acute stroke admissions. This is regardless of time of onset, or time of arrival, or expected outcome.

Clinicians and hospital administrators need to be aware of two forthcoming changes that will likely impact the recording of NIHSS data at the hospital level:

  1. Inclusion of NIHSS score data as an ICD-10 data field.
  2. Proposed stroke performance measure of NIHSS documentation on arrival.

Medical Research: What recommendations do you have for future research as a result of this study?

Dr. Reeves: NIHSS documentation is improving but this study includes only the hospitals who participate in the GWTG-Stroke registry. We need to study NIHSS documentation at all US hospitals who admit stroke patients.

Future research needs to evaluate the success of the forthcoming inclusion of NIHSS score data as an ICD-10 data field. This should be the solution to the fact that NIHSS is missing from administrative (billing) data, but we need to document how complete and accurate this data is.

There are also plans to include recording of NIHSS at admission as a stroke performance measure, which should also increase the completeness of NIHSS data. But again we need to document the rates of compliance with this measure and how accurate the data is.


Variation and Trends in the Documentation of National Institutes of Health Stroke Scale in GWTG-Stroke Hospitals

Mathew J. Reeves, Eric E. Smith, Gregg C. Fonarow, Xin Zhao, Michael Thompson, Eric D. Peterson,Lee H. Schwamm, and DaiWai Olson

Circ Cardiovasc Qual Outcomes. 2015;8:S90-S98,doi:10.1161/CIRCOUTCOMES.115.001775

[wysija_form id=”5″]
Mathew J. Reeves BVSc, PhD, FAHA (2015). NIHSS Stroke Database is Incomplete and May Have Selection Bias 

Last Updated on October 29, 2015 by Marie Benz MD FAAD