Multicenter study of risk factors of unplanned 30‐day readmissions in pediatric oncology

Abstract Background Pediatric oncology patients have high rates of hospital readmission but there is a dearth of research into risk factors for unplanned 30‐day readmissions among this high‐risk population. Aim In this study, we built a statistical model to provide insight into risk factors of unplanned readmissions in this pediatric oncology. Methods We retrieved 32 667 encounters from 10 418 pediatric patients with a neoplastic condition from 16 hospitals in the Cerner Health Facts Database and built a mixed‐effects model with patients nested within hospitals for inference on 75% of the data and reserved the remaining as an independent test dataset. Results The mixed‐effects model indicated that patients with acute lymphoid leukemia (in relapse), neuroblastoma, rhabdomyosarcoma, or bone/cartilage cancer have increased odds of readmission. The number of cancer medications taken by the patient and the administration of chemotherapy were associated with increased odds of readmission for all cancer types. Wilms Tumor had a significant interaction with administration of chemotherapy, indicating that the risk due to chemotherapy is exacerbated in patients with Wilms Tumor. A second two‐way interaction between recent history of chemotherapy treatment and infections was associated with increased odds of readmission. The area under the receiver operator characteristic curve (and corresponding 95% confidence interval) of the mixed‐effects model was 0.714 (0.702, 0.725) on the independent test dataset. Conclusion Readmission risk in oncology is modified by the specific type of cancer, current and past administration of chemotherapy, and increased health care utilization. Oncology‐specific models can provide decision support where model built on other or mixed population has failed.


| METHODS
A look-back period of 6 months from the index/current visit was used to capture data on prior healthcare resource utilization (such as previous visits, previous readmissions). All ICD-9-CM codes were converted to ICD-10-CM for identification of classes of diseases/conditions. We chose classes of diseases according to the chapters of ICD-10-CM codes such as conditions affecting digestive (K00-K95) and genitourinary (N00-N99) systemsall non-neoplastic conditions/comorbidities considered are listed in Table 1.
An a priori threshold of 100 responses/encounters was selected to guide against absolute rarity or sparsity of variables/covariates and to reduce the potential for problems (such as inestimable or inflated effect sizes) due to statistical separation during model development. 14 The level of multicollinearity was assessed by estimating the generalized variance inflation factor (GVIF) across the candidate variables [15][16][17][18][19] and excluding variables with GVIF greater than 4 as a threshold. 19 Statistical separation and multicollinearity can lead to unstable models and false discoveries, hence the need to address them before model development. Summary statistics of all variables considered during model development are shown in Table 1.  Table 1)   cancer medication given increases the odds of readmission by 14.6%.

| RESULTS
The maximum number of unique cancer medications administered to a single patient in this study was 9 with a mean and median of 1 cancer medication per patient in the study. This does not address the difference in risk that may exist by the type of cancer medication administered.
The effect of the statistical interaction between Wilms Tumor and the administration of chemotherapy is shown in Figure 2 The area under the precision-recall area under the curve was calculated to be 0.614.

| DISCUSSION
Predictive modeling for pediatric oncology hospital readmissions has largely been left unaddressed in medical and data science literature.
These patients are, however, universally seen as patients at the highest risk for readmission (as captured in existing models for the general pediatric population). 6,11,12 The challenge with general pediatric readmissions models is that the models often suffer from poor specificity in stratifying risk in oncology.
Most findings in this study include unmodifiable risk factors, as There are also elevated odds for readmission among patients with ALL in relapse, bone and cartilage cancer, neuroblastoma, and rhabdomyosarcoma. Patients who are in relapse for ALL are often in need of a strong reinduction of chemotherapy. 23,24 Additionally, these patients may be sick at the time of relapse, and therefore require more clinical attention and resources than newly diagnosed ALL patients. Neuroblastoma and rhabdomyosarcoma also showed an increase in odds of readmission. These cancer types are known to have aggressive metastasizing behaviors. 25 Additionally, similar to the treatment of higher-risk leukemia, neuroblastoma is known to be treated with high-risk transplant procedures in addition to chemotherapy, surgery, and radiation. 25 Each of these procedure types is associated with immediate and long-term risk factors for infection, myopathy, and other complications. 23,25,26 Rhabdomyosarcoma is the third most common extracranial solid tumor in pediatrics and one of the cancers for which prognosis for advanced stages and those harboring deleterious genomic abnormalities has not improved. 27 Our findings indicate the need for improved quality of care of these patients to reduce unplanned readmission after discharge from the hospital. Brain cancer, conversely, showed a reduction in odds of readmission. This reduction in risk makes clinical sense because brain cancer, when operable, is often treated with surgery and chemotherapy sessions that are not as frequent and intensive. 28 A common complication from the treatment of cancer is chemotherapy-induced neutropenia, which has been reported as a cause of readmission. [29][30][31] The treatment of chemotherapy-induced neutropenia is, however, complex and may vary considerably across cancer types, aggressiveness of treatment, providers, and institutions.

ETHICAL STATEMENT
This study was approved by the Children's Hospital of Orange County's Institutional Review Board with IRB number 180857. Patient consent was waived by the IRB considering the de-identified nature of the database.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on reasonable request from the corresponding author, approval from the corresponding author's Institutional Review Board, and approval from Cerner Corporation. The data are not publicly available due to privacy or ethical restrictions.