Effect of electronic patient record use on mortality in End Stage Renal Disease, a model chronic disease: retrospective analysis of 9 years of prospectively collected data

Care of chronic disease patients now predominates in medical practice, and accounts for >75% of US $2.1 trillion medical care costs [1]. Health information technology is believed essential in improving outcomes and decreasing costs [2, 3]. Although an integrated electronic medical record, performance measures, and active performance monitoring has been associated with improved quality measures [4–6], hopes for effects on healthcare quality, efficiency, and costs have yet to be realized. One recent systematic review concluded that the evidence was sparse [7], another found little evidence that computerized clinical decision support has effected patient outcomes [8]. We found no literature on effects of electronic medical records on mortality in any chronic disease.

Over 30 years ago Fries concluded that in chronic disease: "a major failure of the traditional chart is its inability to indicate adequately complex temporal relationships between clinical, laboratory, and therapeutic events [9]." To address these complex temporal relationships, paper spreadsheets were used to manage clinical, laboratory, histological, and therapeutic details sequentially in renal involvement in systemic lupus erythematosus [10]. This approach was expanded to develop a new comprehensive patient-centered paper record to facilitate understanding of systemic and renal manifestations of diseases. After extensive testing it was converted to electronic form [11]. Discrete medical practice details for diseases affecting multiple organ systems were transformed into coded data, enabling rearrangement of data elements to facilitate clinical practice and observations over many years.

End Stage Renal Disease (ESRD) is complex, costly, and affects more than 500,000 patients in the United States [12]. As a model to test effects of this comprehensive, coded and analyzable, electronic patient-centered record (EPR) in chronic disease, ESRD has unique advantages. First, it is highly regulated, with frequent routine and ad hoc State and Federal oversight; treatment adequacy markers must be reported. Second, to start dialysis treatment, patients must meet mandated criteria. Third, treatments, devices, and drugs used are largely standardized. Fourth, dialysis procedures are also largely standardized and, for the most part, operator independent. ESRD often results from primary disease conditions such as diabetes mellitus, HIV/AIDS, multiple myeloma, systemic lupus erythematosus, arteriosclerosis, and hypertension; comorbid conditions affecting many organ systems are universal. Each patient requires lifetime care, receives 140–156 treatments and accumulates up to 11,600–21,000 individual data items yearly [13], is cared for in many locations by caregivers from several disciplines. Mandatory reporting ensures that the United States Renal Data System provides a national database and mortality standard with which to compare results. Moreover, 68–70% of US patients are treated in dialysis units owned by 5 (since recent mergers, 3) large dialysis "chains" [12]; each deploys an electronic medical record [14–17].

We here test the hypothesis that "Successful management and treatment of the patient and the important individual manifestations of a chronic disease require complex feedback systems that relate therapeutic interventions to clinical and laboratory information relevant to multi-organ systems over prolonged periods" [11]. We analyzed retrospectively data collected during patient care from 1998 onward using the EPR, enhanced continually since 1976 [13, 18–21]. This study EPR was the test instrument. USRDS mortality data provided contemporaneous controls [12]. Favorable effects on mortality and cost were observed on ESRD patients treated by chronic maintenance hemodialysis (HD) for a total of 3,924 years over a 9-year period.

Using an earlier EPR version in a single dialysis unit, we reported mortality 25% below the 1980–1989 national average [13, 22]. Both EPR software and the national database have been greatly enhanced since. The present study is the first demonstration, to our knowledge, that use of an electronic patient record in any chronic disease practice favorably effects mortality.

Stead wrote recently that improvements from information technology implementation "are difficult to quantify in a practice while changing people's roles, process, and technology at the same time. Most measures have an immediate impact on process whereas many of the expected benefits are in long-term clinical outcomes [8]." We had no a priori knowledge of time needed from initial EPR implementation to development of a measurable outcome, particularly when dependent on many underlying diseases, comorbid factors, care, and other variables. A favorable effect by year 3 seems reasonable because staff had to learn to use and become comfortable with the technology while continuing daily patient care activities, and lower mortality in year 3 is likely to reflect improved quality of care in year 1 and/or 2. This was clearly evident in Units B and C by year 3, and in Unit A by year 5. Lower mortality was observed consistently thereafter for 5 years in Unit A, 6 years in Unit B, and 5 years in Unit C.

That mortality of the study population was less than USRDS might be explained by differences between demographic and comorbid factors of the study and USRDS populations. To test for this possibility we obtained standardized mortality ratios (SMR) generated by the University of Michigan Kidney Epidemiology and Cost Center [23]. Since 2001, SMR has been calculated from a Cox model, adjusting for age, race, ethnicity, gender, diabetes, ESRD duration, patient comorbidities and body mass index at incidence, and population death rates. SMR trends in 2001–2006 were similar to those in our analysis. Compared on a year-by-year basis, SMR and the study patient mortality correlated significantly (adjusted r² = 0.30, p = 0.011), suggesting that the decreased mortality in 2001–2006 was not due to differences in demographic and comorbid factors of the study and USRDS populations. The demographic and comorbid factors of the population of each Unit, which varied year-to-year almost imperceptibly, cannot explain the lower mortality observed in each Unit. This supports the view that the lower mortality when compared with USRDS was not due simply to a center effect.

We sought to identify other factors such as change in management patterns or dialysis technology that might have contributed to improved outcomes; no obvious ones were found. Increased number and/or professional qualification of patient care staff is one possible explanation; this was not the case. As in an earlier report [13], there was a favorable effect on staff efficiency (Table 7). Compared with years 1–2 hemoglobin and serum albumin, indicators of patient well being, changed little: hemoglobin increased 5% in years 3–9; serum albumin and serum iron were unchanged. Kt/V, a widely used index of dialysis adequacy, increased in years 3–9 by 11%, 2%, and 4% in Units A, B, and C respectively. Mortality decreased most in Units B and C, where the change in the index was very small.

How is it possible that the particular study EPR can have played so important a role? Two properties of the EPR are crucial to understanding: it is patient-centered, and it is extensively coded.

The patient-centered study EPR captures, stores, and retrieves on-line and without delay all patient-specific medical data from multiple information domains including diagnoses, procedures, symptoms, signs, medications, orders, test results, and dialysis treatments [11, 18]. It does so for all venues of care and for care provided by any caregiver, especially important in dialysis where all patients have multiple systemic diseases that need repeated evaluation and treatment by multiple providers over many years at a variety of sites. Unlike disease registries that usually focus on a single disease entity and its assumed complications, the study EPR is a generalized model of medicine that makes no assumption about future co-morbidities, complications, or outcomes in its data model. On the contrary it captures, stores, and retrieves any or all that might occur.

Extensive coding of the study EPR is essential to its utility. The ultimate products of virtually all electronic medical records are notes similar to those in a paper chart; data cannot be added or rearranged to explore varied and unexpected facets of the patient's condition(s), nor can they be viewed over time. The products of the study EPR are reports that contain many domains of data appropriate to a particular disease at any particular moment in time in care of the patient (Figures 1, 2, and 3). Their function is not static documentation of a moment; rather it is to display what has occurred over time, to evaluate and/or change the intervention, and then repeat the process. Essentially the report facilitates continuous quality improvement at the point-of-care [24].

These properties of the study EPR made possible:

(1) Computerized data entry with, and generation of, clinically relevant reports thereby eliminating communication and process errors.

(2) All information needed for skilled clinicians to make best judgments about patient care was always available immediately in clinical reports. Data in reports were complete, from multiple domains, and were viewed sequentially over time (Figure 3). We believe that complete, accurate, up-to-date, and focused time-oriented information had a favorable effect on work flow, facilitated understanding of the complex interrelations that effect outcome, enabled clinicians to make best management decisions, and therefore favorably impacted survival.

(3) Aberrant findings in one or a few patients were easily recognized (Figure 7), and their prevalence investigated rapidly [25, 26]. This knowledge led to development of reports (Figure 8) and rules based alerts and reminders (Figure 9) for use in daily clinical practice. Knowledge so derived also enabled survival analysis [27], generated new understanding of a disease [25], decreased symptoms, improved well being [13, 18, 22], and detected and solved unexpected problems [19].

(4) Development of protocols and their timely updating as new data became available. Recognition of a high prevalence of iron deficiency in anemic epoietin treated dialysis patients led to systematic study, with repeated feedback, of intravenous iron repletion. Unexpected findings included: a very significant rise in serum albumin – an excellent outcome marker [28] – and in indirectly measured muscle mass; iron deficiency without iron deficiency anemia; decreased hospital admissions and stay [26]. This knowledge was disseminated in the study units when one of us (JAL) joined the Rogosin staff.

A trial of the study EPR with a randomly selected control group has the potential to provide evidence of cause and effect but would be difficult to design and control, and expensive to carry out for the many years needed. The present study compares data between groups and across time periods and uses two national standards, USRDS data [12] and standardized mortality ratio data [23], for contemporaneous controls. Unadjusted and adjusted mortality decreased comparably in 3 practice environments using the study EPR. Nationwide, where 70% of patients are treated by 3–5 corporations that use a traditional electronic medical record focused on management and billing, there was no concurrent change. With the limitations inherent in an observational study, the data provide evidence that benefits in long-term clinical outcomes can be achieved with the study EPR although cause and effect cannot be proven.

The study EPR had a favorable effect on mortality in ESRD patients treated by dialysis. Are similar results possible in other chronic diseases, and with other medical information systems? The challenges are great, as reviewed recently in a study of practice guidelines and quality of care for older patients with multiple comorbid conditions [29]. In 1977 we wrote, "At any time in the course of an illness, the patient may develop one, or many of the universe of symptoms, physical signs, and laboratory abnormalities that occur in clinical medicine as a whole" [11]. The study EPR database, and its integrated query tools, were designed to enable deployment in many fields of medicine and chronic diseases. If this and other EPR databases are to yield comparable results, we believe it essential that they adhere closely to the dictum that the medical record must be able to indicate complex temporal relationships between clinical, laboratory, and therapeutic events [9], and in both individuals and groups of patients.