Evaluating Consistency of Documented ICD Codes for Hospitalized Patients With Sickle Cell Disease

This is a secondary analysis of hospital admission claims, for both principal and secondary diagnoses, occurring between April 2015 and December 2016 in 4 states (California, Florida, New Jersey, Pennsylvania). Claims data provided information on ICD, Ninth Revision, and ICD, 10th Revision, diagnosis codes, comorbidities, and demographics. Linked data from the American Hospital Association Annual Survey provided information on hospital characteristics.

The sample included patients hospitalized at least 2 times during the study period, with at least 1 hospitalization including a diagnosis code for SCD or SCT (ICD, Ninth Revision [282.X], ICD, 10th Revision [D57.X]). Patients were included if they were aged 0 to 99 years and were hospitalized in a facility represented in American Hospital Association data. Hospitals with <10 records of patients with SCD/SCT were excluded.

Overall, 80 031 hospital records belonging to 15 380 unique patients with documented SCD/SCT were identified (Table 1, Sample A). Sample A was subset into 2 samples: Patients with a diagnosis code for SCD only (Table 1, Sample B), and patients with a diagnosis code for SCT only (Table 1, Sample C). There were 643 patients who had both SCD and SCT coded. In these cases, patients were considered SCD only because, clinically, these conditions are mutually exclusive. Further, the documentation of SCT was coded as “missing” and patients were categorized as being “not consistently documented.”

Patient demographics included age, sex, race/ethnicity, and type of insurance. Hospital teaching status was measured as the ratio of physician residents/fellows to beds. Hospitals were categorized as nonteaching (having no residents/fellows), minor (≤1:4 residents/fellows to beds), or major (>1:4 residents/fellows to beds). Bed size was classified as small (<100 beds), medium (101–250 beds), or large (>250 beds). A hospital with high-technology capabilities included hospitals with the ability to perform open heart surgery and/or major organ transplants. Geographic location (ie, rural/urban) was determined by the metropolitan statistical area. Hospitals were categorized as either a general acute care hospital or a freestanding pediatric hospital.

For each sample, patients were classified on the basis of whether they had a SCD/SCT diagnosis code documented in every hospitalization record (ie, consistently documented). If the patient was not consistently documented, they were defined as not consistently documented, meaning that at least 1 hospitalization record lacked documentation of their known SCD/SCT. Descriptive statistics were used to report demographic differences between consistently and not consistently documented patients. For Table 1, Sample A, we report the top 5 reasons for hospitalization (excluding SCD/SCT as a primary reason) among patients with and without a document SCT/SCT on the hospitalization.

Hospitals were rated as either “better,” “average,” or “poor” on the basis of how frequently a hospital had hospitalization records without a documented SCD/SCT code. Better hospitals were those with fewer hospitalization records without a documented SCD/SCT diagnosis ≥1 SD below the mean. Poor hospitals were those with hospitalization records more often had a not documented SCD/SCT diagnosis ≤1 SD above the mean. Descriptive statistics describe differences in hospital characteristics among hospitals with better, average, and poor ratings. Analyses were conducted using Stata. χ² tests of significance were performed.

Table 1 reports patient demographics for the sample of patients with SCD/SCT (Sample A), and its subset samples: patients with SCD only (Sample B) and SCT only (Sample C). The samples are further disaggregated by whether the patient’s conditions were consistently documented. Forty percent of patients with SCD only were not consistently documented, compared with 78% of patients with SCT only. Among those with SCD only who were not consistently documented, patients tended to be older (35% 65+ years [P = .0002]) compared with the SCT only groups, where the largest proportion (39%) of not consistently documented patients were those aged 18 to 34 years [P < .001]. Black patients were more likely to have their condition not consistently documented in the SCT-only group as compared with the SCD-only group (86% [P < .001] vs 45% [P < .001]). Overall, SCT-only patients were more frequently not consistently documented compared with SCD only (78% vs 40%).

The top 5 reasons for hospitalization (excluding SCD/SCT as a primary reason) among patients with and without a documented SCD/SCT on the hospitalization record are presented in Table 2. Septicemia, pneumonia, and acute kidney failure were common among both groups. This may suggest patients with SCD/SCT are hospitalized for reasons related to their underlying SCD/SCT, and yet, we find their SCD/SCT is not documented during that hospitalization.

Differences in hospital characteristics among hospitals with better, average, and poor ratings in SCD/SCT documentation are described in Table 3. Most hospitals were in California (43%) and Florida (30%). California had more hospitals in the poor category compared with the better, and Florida had twice as many hospitals in the better category than in the poor for Samples A and B, with negligible differences in Sample C. Major teaching hospitals (29%) were more likely to be better hospitals compared with nonteaching (17%) [P = .040] in Samples A, similarly in Sample B, but not in Sample C, where major teaching hospitals were less likely to be better hospitals compared with nonteaching (7% vs 16% [P < .001]). There were no pediatric hospitals in the poor category. Small hospitals were more likely to be poor (23%) than large hospitals (14% [P = .027]). This was especially true in Sample C (45% vs 9% [P < .001]). In general, similar trends were consistent across subset samples.

We identified that about half of patients with SCD/SCT do not have their condition consistently documented across all of their hospital records. Subset analyses which separately analyzed documentation consistency among patients with SCD versus SCT revealed that not having consistent documentation was more prevalent among individuals with SCT. This is a novel contribution to the broader literature regarding documentation accuracy in this population. As research on SCT is expanding,^11,12  accurate documentation is especially salient for this population.

Another novel finding of this analysis is the investigation of the top reasons why patients with SCD/SCT are admitted to hospitals, and the similarities in those reasons regardless of whether the patient’s SCD/SCT was documented. We found that common comorbidities associated with SCD,^1,6,13,14  such as septicemia, kidney failure, and pneumonia, were similar reasons for hospitalization among both groups of patients. From a research perspective, these findings highlight limitations of administrative data that exclusively rely on ICD coding. Utilizing multiple hospitalization records may enhance researcher’s ability to more accurately study health outcomes of patients with chronic conditions like SCD/SCT.

Regarding hospital ratings, there were similarities and differences across subsamples identifying that certain hospitals do better at documenting SCD/SCT, whereas other hospitals have room for improvement. Pediatric hospitals were consistently better hospitals, although that may be attributed to the small number of pediatric hospitals in our sample (n = 13), or because of advocating parents, recent diagnosis, and follow-up. Although both California and Florida are among the states with the highest rates of SCD Medicare, Medicaid, and Children’s Health Insurance Program beneficiaries,^2,15  they do not share the same distribution of better- or poor-rating hospitals, with Florida hospitals performing better. This is surprising because California has a longer history of participating in national SCD surveillance programs.^16,17  Identifying what documentation processes are different between these 2 states may help in identifying interventions to improve SCD/SCT documentation.

We analyzed cross-sectional data using descriptive statistics and hypothesis tests; therefore, causal relationships cannot be inferred. Data did not differentiate “observation” and “inpatient” stays; and thus, we were unable to interrogate the hypothesis that coding may be sparser for patients in observation stays. Future research should expand on documentation of clinical encounters outside of the hospital, such as in primary care.

Inconsistent documentation is prevalent, and more so among patients with SCT than those with SCD. Our findings have broad implications for research on many chronic diseases that rely on administrative claims data to evaluate prevalence, treatments, and patient outcomes. Future research that depends on administrative claims data to study SCD/SCT patients should attempt to identify all hospitalization records for a patient. Further research is needed to determine what makes certain hospitals better at consistent documentation.