Placing clinical variables on a common linear scale of empirically based risk as a step towards construction of a general patient acuity score from the electronic health record: a modelling study

This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the license. See: http://creativecommons.org/licenses/by-nc/3.0/ and http://creativecommons.org/licenses/by-nc/3.0/legalcode

Quantitative or categorical clinical variables can be transformed into risk functions that correlate well with in-hospital risk. This methodology provides an empirical way to assess inpatient risk from data available in the Electronic Health Record. With just the variables in this paper, we achieve a risk score that correlates with discharge disposition. This is the first step towards creation of a universal measure of patient condition that reflects a generally applicable set of health-related risks. More importantly, we believe that our approach opens the door to a way of exploring and resolving many issues in patient assessment.

Pearson correlation coefficients comparing in-hospital risks with postdischarge risks for creatinine, heart rate and a set of 12 nursing assessments are 0.920, 0.922 and 0.892, respectively. Correlation between postdischarge risk heart rate and the Modified Early Warning System (MEWS) component for heart rate is 0.855. The minimal excess risk values for creatinine and heart rate roughly correspond to the normal reference ranges. We also provide the risks for values outside that range, independent of expert opinion or a regression model. By summing risk functions, a first-approximation patient risk score is created, which correctly ranks 6 discharge categories by average mortality with p<0.001 for differences in category means, and Tukey's Honestly Significant Difference Test confirmed that the means were all different at the 95% confidence level.

No multivariate analysis was performed on the example variables, making the associations found subject to possible unknown confounders; also, the work has been done at a single site with a population skewed older than the general population.

A large dataset (more than 40 000 hospital visits) was used to derive the risk functions; this is a new empirical method for evaluating univariate risk, independent of diagnosis or comorbidity, and without using population norms or expert opinion.

Risk functions are easily computed with the data from an Electronic Health Record and the Social Security Administration Death file; these functions correlate well with in-hospital mortality, giving investigators a new tool to study the acuity of patients in the hospital.

Introduction

Clinicians regularly utilise various systems designed to quantify some aspect of patient acuity.1 In most cases, these assess risk pertaining to: a specific event such as cardiopulmonary arrest or transfer to intensive care2–11; or to a specific disease or procedure12–16; or within a specific environment, such as the intensive care unit (ICU)15 17 18 or for after-the-fact risk adjustment, such as to compare the performance of medical units.19–22 However, there has been no previous system created to score the real-time overall condition of individual patients within a hospital's general ward, across the acuity spectrum, based upon empirical evidence from the Electronic Health Record (EHR).

In this study, we lay the foundation necessary for an overall measure of a patient's condition. We seek to create a contemporaneous longitudinal index, calculated by summing empirical estimates of incremental risk. Systems to measure risk in the hospital have been based on aggregated expert opinion,15 18 23–27 or regression models.17 28 And, for laboratory tests, risk as reported to physicians by pathology labs is usually based on the norm of a ‘healthy’ population29 with the notion that if a measurement is within the reference range (mean ± 2 SD), there is no risk. Unfortunately, this lab method has no direct link to risk; for example, serum cholesterol for the adult population would have placed the norm at 200 mg/dl,30 which in light of extensive medical evidence is now understood to be ‘borderline high’.31

We introduce a different method to estimate a patient's risk, which does not rely on or require expert opinion, or a regression model, or a population norm, but rather is completely empirical and evidence based. Our hypothesis is that placing clinical variables on a linear scale of all-cause postdischarge mortality produces risk functions that are directly correlated with in-hospital mortality. Adding together the risk functions of differing underlying metrics is a step towards creation of a general patient-condition score of empirically based risks. These functions are readily computed by combining clinical data available in a hospital's EHR with mortality data available from the Social Security Administration.

This is one of a series of studies whose objective was to demonstrate and validate the creation of such an index, derived empirically from regularly collected variables available in a hospital's EHR. In a previous study, we demonstrated that for nursing assessments, predischarge assessments are strongly correlated with 1-year postdischarge mortality, and nursing assessments at admission are correlated with in-hospital mortality.32

We extend this work in three ways: first, by computing risk functions for vital signs and for laboratory blood tests; second, showing the relevance of 1-year postdischarge risk functions to risk in the hospital, by computing the correlation between in-hospital risk and postdischarge risk and third, showing that the sum of risk functions correlates with patient acuity at the time of discharge, as represented by the patient's discharge disposition (eg, to home or rehab or a skilled nursing facility).

This common linear scale of a risk function reflects the health consequences of any value of the variable in terms of all-cause risk of mortality associated with that value, independent of diagnosis. One advantage of having various routinely available in-hospital clinical variables expressed in terms of per cent risk is that they then can be linearly added in some fashion to assign a total risk index for each patient at any moment in time, given the variable values for that patient. The current study illustrates our new methodology using several basic variables as an example, including quantitative, such as heart rate or creatinine level, and categorical, such as 12 pass/fail nursing assessments. We then demonstrate the utility of a first-approximation risk score based on this example, which we compute by simply adding the risks associated with these example variables.

The full details of construction and validation of a real-time, inpatient condition score are the subject of a forthcoming study. This new measure is currently being used and evaluated in several medical centres, and is called the Rothman Index in memory of Florence A. Rothman, whose death inspired this research. The various measures necessary to form an index in other areas of research can be determined by the methodology developed here, and we encourage application of our methods.