Determining causes and potential preventability of clinical deterioration among oncology inpatients

Patrick Lyons, M.D.


Hospitalized patients with cancer face unacceptable risk for clinical deterioration. With more than twice the risk as the general inpatient population, more than 9% progress to the intensive care unit (ICU) or die. Many of these patients would have opportunities for cures or long remissions if not for intercurrent clinical deterioration due to treatment- or disease-related complications. Unfortunately, no studies have evaluated the causes or potential modifiability of clinical deterioration among hospitalized patients with cancer. This proposal aims to create knowledge of subphenotypic paths into deterioration in hospitalized cancer patients, under the scientific premise that early identification of these subphenotypes can improve outcomes through tailored evidence-based interventions. Specifically, the proposal aims to quantify (a) the scope of the problem – clinical deterioration in oncology patients – and (b) the prevalence of preventability among these patients; and to characterize these patients, their cancer-directed treatments, and their deterioration events, such that important links between clinical characteristics and outcomes can be identified. Elucidating these patterns is an important step towards sophisticated predictive analytics approaches to pre-empt deterioration.


Project Summary:

The purpose of my work is to create, implement, and test a systematic program (an “Early Warning System,”
EWS) to optimize clinical outcomes for hospitalized patients with cancer by anticipating and intervening on
clinical deterioration (i.e., getting sicker and requiring a higher level of care). Many of these patients would
have opportunities for cures or long remissions if their cancer treatment were not interrupted by clinical
deterioration due to treatment- or disease-related complications. Standard EWS protocols currently produce
many false alarms for the general inpatient population while failing to accurately identify oncology patients at
risk. Thus, my colleagues and I designed the first oncology-specific EWS (the OncEWS), which displays a
higher positive predictive value and greater accuracy than existing EWS.

However, important gaps exist in this area’s current research. Most urgently, identifying high-risk patients is
not enough to improve outcomes. Patients can benefit from an EWS if and only if (a) their deterioration has a
fixable cause and (b) clinical teams quickly provide the right treatments for these causes. Promptly delivering
the right treatment requires correctly identifying the issue responsible for decline. Unfortunately, no prior
research has described these causes of deterioration, nor have any studies looked at the potential modifiability
of clinical deterioration among hospitalized patients with cancer.

In this application, I will address these gaps by (1) quantifying the relative frequencies and potential
preventability of important causes of clinical deterioration (e.g., sepsis, respiratory failure) among oncology
patients who experienced clinical deterioration, and (2) determining the extent to which cancer treatments are
related to clinical deterioration.

In the first Aim, I will oversee chart reviews from 800 hospitalized patients with cancer who experienced
clinical deterioration. This Aim will determine the relevant clinical diagnoses responsible for deterioration and
the likelihood that each case to have been prevented, had earlier recognition of a problem (and subsequent
action) occurred.

In the second Aim, I will document whether oncology patients who deteriorated received cancer-directed
treatments during the weeks immediately preceding their deterioration, and make determinations as to the
relationship, if any, between oncology treatment and deterioration.

These aims will allow me to (1) accurately describe the scale and scope of the problem (potential
modifiability of deterioration in hospitalized patients with cancer via the systematic approach I describe above),
(2) use this information to train “bespoke” machine learning models to more accurately predict these
deterioration syndromes, (3) understand ways in which cancer treatments contribute to – and are interrupted
by – clinical deterioration, and (4) move towards implementation and clinical testing of an accurate EWS for
oncology patients (i.e., by using this estimation of “effect size” to design a clinical trial).