Development of an electronic health record-based chronic kidney disease registry to promote population health management

Setting

PHS is a non-profit hospital and provider network based in eastern Massachusetts, serving close to 6 million patients. Brigham and Women’s Hospital (BWH) and Massachusetts General Hospital (MGH) are the flagship tertiary referral academic centers in the organization, with highly specialized medical and surgical services offered. Several traditional community medical centers provide general inpatient services. Several outpatient practices spread throughout the greater Boston area are part of PHS. Starting in 2015, a common EHR has been used by ambulatory and inpatient providers across the network. We organized a system-wide CKD collaborative, comprised of nephrologists and Primary Care Providers (PCPs). This collaborative has provided oversight of the registry development.

Patient identification for inclusion in CKD registry

Figure 1 depicts inclusion criteria for the CKD registry, including laboratory data, diagnosis codes, and visit data. Patients are included in the CKD registry first by identifying alive patients at least 18 years old that are ‘active’ in our system, defined as having a prior ambulatory or inpatient encounter within the previous 5.5 years or a future scheduled encounter; as well as having recently updated wellness information such as vaccinations or laboratory values.

Among these active adult patients, we include patients if they meet one of the following criteria: 1) most recent eGFR < 60 ml/min/1.73m² and one additional eGFR at least 90 days prior < 60 ml/min/1.73m², with both values recorded within the last 3 years;or 2) at least two values of urine total protein or urine albumin > 300 mg/gm; or 3) ESRD or dialysis on EHR problem list or as an International Classification of Diseases (ICD)-10 diagnostic code during an ambulatory or inpatient encounter. Prior to March 2018, the EHR calculated eGFR using the Modification of Diet in Renal Disease (MDRD) equation. Subsequently, the EHR has utilized the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) serum creatinine equation, adjusting for African American race by multiplying generic eGFR values by 1.212.

Our CKD registry includes 60,503 patients as of July 31, 2018. The registry updates data in real-time, so the total number of patients included varies daily. The vast majority of patients (98%) are identified via eGFR values, and 91% have a PCP within Partners (Fig. 1).

CKD and 5D staging

Our CKD collaborative developed the algorithm for classifying patients in stages 1 through 5 (Fig. 2). Given that eGFR can fluctuate for a patient over the course of several years, there are multiple approaches to such classification, including ‘average eGFR’, ‘lowest eGFR’, and ‘most recent eGFR’. We opted to base classification on the most recent eGFR to reduce confusion among front line clinicians, recognizing the limitation that this may sometimes capture acute kidney injury.

We classified patients as 5D (requiring dialysis) or ESRD based on the presence of active dialysis ICD-9 or ICD-10 codes (including such common codes as ICD-9 CM: V56.0, V56.1, 585.6, V45.11; and ICD-10 CM: N18.6, Z99.2) or recent outpatient dialysis treatments regardless of eGFR (see Additional file 1: Table S1). We classified patients as stage 5 if no criteria for active dialysis is met and the most recent eGFR is < 15. Patients are classified as ‘no stage assigned’ if there are no eGFR values in the last 3 years and no active dialysis criteria are met.

CKD registry metrics

Data management

The CKD registry is available for use by individual clinicians managing their own patient panel as well as clinical directors overseeing an entire CKD population. The data collected within the EHR registry are backed up nightly to a separate enterprise wide data warehouse, where the data can be combined with external data sources such as claims and cost data to facilitate more advanced analytics. Only approved users within the PHS network who have been trained on appropriate use and stewardship of data are permitted to access and query the EDW to obtain aggregate data across providers and clinics.

Registry data validation

We validated the registry data using two sequential steps. First, 100 patients were randomly selected from each of stage 4, 5, and 5D designations as determined by our CKD staging algorithm (Fig. 1). Two study authors (SA and MFM) performed chart reviews to assess for appropriate staging, the presence or absence of arteriovenous fistula or arteriovenous graft (AVF or AVG), and transplant status. This first step allowed refinement of our CKD staging algorithm for 5D patients, such as adding logic to capture inpatient dialysis progress notes.

Second, we randomly identified an additional 300 patients with stage 4, 5, and 5D CKD for additional manual chart review. We performed additional validation of stage 3 patients. We demonstrated a positive predictive value (PPV) of 100% (95% CI 96.4–100%) for stage 3 and 4 patients, 82% (95% CI 73.1–89.0%) for stage 5 patients, and 91% (95% CI 83.6–95.8%) for stage 5D patients. Nearly all (99%) patients who were classified as stage 5 or 5D were validated as having at least stage 5. However, we found that patients were sometimes classified as stage 5D when not on dialysis, or stage 5 when on dialysis, (11 and 20% of cases, respectively). This could be due to documentation of dialysis treatment only in scanned documents not identifiable in automated data extracts, recent dialysis discontinuation mentioned in free text notes, or reversal of decisions to pursue dialysis.

Our second chart review also demonstrated a PPV of 100, 91 and 83% for vascular access for stages 4, 5, and 5D respectively; and a PPV of 93, 95 and 79% for transplant status. Most discrepancies were due to information contained within the body of progress notes.

Patient demographics and clinical characteristics

Assessment of quality metrics

Vascular access placement

Transplantation status

Slightly more than one-half (54%) of all stage 5D patients have been referred for transplantation (Table 4), and 37% of stage 5 patients have been referred. A small proportion (2%) of patients representing each CKD stage are prior transplant recipients.

Utilization metrics

Opportunities to create value: the identification of gaps in care

Studies to date illustrate the need for regional or network-based CKD registries that aggregate objective data [23], but have not been focused on clear care gaps that can be utilized by both PCPs and nephrologists. Our CKD registry suggests multiple areas for improvement in the delivery of care. With respect to early CKD management often performed by PCPs, rates of proteinuria testing and ACE-I/ARB treatment should be targeted. For advanced CKD and nephrologist-based care, nephrotoxin avoidance, hepatitis B immunization, vascular access placement and transplant evaluation can be improved. CKD patients in our registry are high utilizers of inpatient care, consistent with findings in the general CKD population [6]. The inclusion of the Tangri risk scores and utilization data will help identify patients at greatest risk and need for intervention.

Within our institution, the first phase of CKD registry implementation will involve education of clinicians across PHS practice sites, followed by implementation of a registry coordinator to promote population health strategies.

Population health strategies for CKD management

The CKD registry facilitates data driven population health interventions, whereas previous efforts have been based on national data trends. The goal is to slow the progression of renal disease, prevent complications, and optimize transitions along the CKD spectrum (Fig. 3) [7].Our organization is currently utilizing the CKD registry to deploy population health management strategies including e-consults, computer decision support (CDS) for nephrology referral, PROMs and ESRD care coordination. We are in the process of developing additional CDS for quality metrics, and implementing advanced CKD care coordination. The first challenge is to improve rates of CKD recognition. Our robust specialty e-consult program can facilitate early CKD management by PCPs with nephrology support, without in-person visits. Timely referral of CKD patients to nephrology can be supported via identification of patients at highest risk for progression and CDS. As patients progress through stages of CKD, renal replacement therapy (RRT) planning becomes as important as managing CKD complications. The registry can facilitate coordinated, multi-disciplinary care for advanced CKD by capturing timely, accurate data regarding progression, and the need for vascular access. Ideal application of a CKD registry involves population health coordinators who track performance data over time, and facilitate key care processes. The registry can be leveraged by coordinators to facilitate RRT initiation discussions, modality decisions and conservative or palliative care. Finally, we have incorporated PROMs data in our registry, to drive shared decision-making and gauge patients’ perception of their health. We chose the KDOQL-SF 1.3 to capture PROMS data, because it includes 43 kidney-disease targeted items and has been validated in chronic kidney disease across ethnic groups [24].

Strengths

Our CKD registry has a number of strengths. First, it aggregates data from a large patient population in an integrated network. Second, our registry uses the EHR as a dynamic data source leveraging labs, medication/prescription data, billing, and clinician encounters. Third, our algorithms for CKD stage identification, vascular access, and transplant status were developed iteratively over 2 years based on feedback from PCP, nephrology and population health leadership. Fourth, we have included a broad range of metrics from early to advanced CKD, enabling use by both PCPs and nephrologists. Fifth, the tool is accessible both by individual providers and clinical directors engaged in clinical management as well as population health specialists interested in systems-based interventions. Finally, the registry incorporates important clinical outcomes like CKD progression, ED visits, and hospitalizations that can be utilized to evaluate the impact of interventions on outcomes.

Limitations

Though our CKD registry has undergone multiple iterations before reaching its current state, it may benefit from further revision. Identification and stratification of patients continues to be a major challenge. Identification is inherently limited by screening, and efforts to promote screening are needed. Identification of patients with 5D CKD who receive dialysis at outside dialysis units, but receive inpatient and interventional nephrology care through our system, Our validation approach demonstrates that despite refinement of the 5D identification algorithm, there are limitations to labs, diagnosis codes and targeted word search. This is reflective of the dynamic status of CKD, but also exemplifies fragmented care, specifically RRT, that occurs in disparate locations.

Staging stratification based on eGFR is complex given the moving target in most patients. We define stage based on most recent eGFR for consistency, but the argument could be made for a blended average approach. With regards to loss to follow-up, patients without either a visit or a laboratory test within the past 5.5 years are dropped from the registry to focus care. Future iterations of the CKD registry may include follow-up with patients nearing the time for removal from the registry. Similarly, given that all Partners CKD patients are included in the registry, analysis is limited to positive predictive value and does not include sensitivity or specificity. Another limitation is that the registry is not currently linked to external registries such as USRDS or the Social Security Death Master File, which would ensure more accurate identification of 5D patients and mortality, respectively. Our utilization data is limited given that the registry is restricted to in-network visits, and therefore, may be underestimating total visits. Finally, the low rate of HBV vaccination among stage 5D patients may reflect the fact that they receive most of their care at dialysis units that do not share an EHR with PHS.

Implications

We believe that institutions and networks should adopt EHR-based CKD registries as we move towards value-based care. There is a need for regional and institution-based registries as there is known regional variation in care delivered. Furthermore, as cardiovascular disease remains the greatest cause of mortality in CKD patients, the inclusion of data regarding cardiovascular risk factors in this registry will foster robust multi-disciplinary cardiovascular risk optimization.