Leveraging Health Informatics to Enhance Outpatient Chemotherapy

Operations Management

Majed Hadid

, Adel Elomri

and Regina Padmanabhan

Division of Engineering Management and Decision Sciences, College of Science and Engineering,

Hamad Bin Khalifa University, Qatar Foundation, Doha, Qatar

Keywords: Cancer Care, Data Analytics, Machine Learning, Decision Support.

Abstract: The rise in demand for cancer care services, particularly outpatient chemotherapy, highlights the importance

of improving the management of outpatient chemotherapy operations (OCOM). Despite the numerous studies

addressing OCOM issues, the existing literature has mostly focused on problem-driven research. In this study,

we aimed to utilize data-driven research to identify opportunities for improvement and address research

challenges. To achieve this goal, we collected extensive operational data from a large chemotherapy center

and performed a thorough analysis. Our findings revealed four key research challenges, including the

prediction of length of stay, change in patient drug posting weight, delay in appointment admission, and

stochasticity in drug administration duration. To address these challenges, we developed two machine

learning models to predict these outcomes, utilizing 15 features and highlighting the most important features.

Our results showed an efficient performance in predicting the outcomes using the XGBoost model,

emphasizing the potential of data-driven research in improving OCOM.

1 INTRODUCTION

1.1 The Importance of Outpatient

Chemotherapy Service (OCS)

Outpatient chemotherapy has been increasingly

recognized as a cost-effective and convenient

alternative to traditional inpatient chemotherapy. The

significance of outpatient chemotherapy lies in its

ability to provide patients with the same quality of

care while reducing the burden of hospitalization.

Outpatient chemotherapy was associated with lower

costs, shorter hospital stays, and increased patient

satisfaction compared to inpatient chemotherapy

(Houts et al., 1984). This has important implications

for health care systems as it can free up hospital beds,

reduce wait times, and improve patient outcomes.

Outpatient chemotherapy was not only associated

with improved patient satisfaction, but also with

improved clinical outcomes, as patients received their

treatments in a timely manner without being admitted

https://orcid.org/0000-0002-1542-3220

https://orcid.org/0000-0003-1605-9800

https://orcid.org/0000-0001-9448-6950

to the hospital (Waller et al., 2014). These findings

highlight the importance of promoting and expanding

the availability of outpatient chemotherapy services.

1.2 Operations Management (OM)

Challenges in OCS

However, managing the operations of outpatient

chemotherapy process presents several challenges.

One major challenge is managing patient flow and

ensuring that patients receive their treatment in a

timely and efficient manner (Lamé et al., 2016).

Additionally, managing inventory and ensuring that

the right drugs are available at the right time is

another significant challenge (Hadid et al., 2021).

Moreover, managing the complex and dynamic

nature of the outpatient setting, including

unpredictable patient volume and the need for rapid

adjustments to accommodate for changes in patient

needs, can pose significant challenges for outpatient

chemotherapy providers.

378

Hadid, M., Elomri, A. and Padmanabhan, R.

Leveraging Health Informatics to Enhance Outpatient Chemotherapy Operations Management.

DOI: 10.5220/0012348600003657

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 2, pages 378-384

ISBN: 978-989-758-688-0; ISSN: 2184-4305

1.3 Gaps in Outpatient Chemotherapy

Operations Management (OCOM)

Despite the recognition of the importance of

outpatient chemotherapy and the challenges

associated with its operations management, there is

still a gap in the literature related to effective

strategies for managing the operations of outpatient

chemotherapy facilities (Evans et al., 2016). The

limited research has primarily focused on patient flow

and wait times, rather than on the broader operations

management issues associated with outpatient

chemotherapy (Lamé et al., 2016).

There is also a lack of research on the impact of

patient characteristics and behavior on the efficiency

and effectiveness of outpatient chemotherapy opera-

tions (Ahmadi-Javid et al., 2017). Additionally, there

is limited research on the use of advanced analytics and

technology to improve the management of outpatient

chemotherapy operations (Mandelbaum et al., 2019).

As technology continues to play a larger role in

healthcare delivery, it is important to understand how

it can be leveraged to improve the management of out-

patient chemotherapy programs and support patients in

their care. Although the use of data has the potential to

greatly enhance decision-making in the operations

management of outpatient chemotherapy, there has

been limited research on the application of data

analysis techniques in this field (Hadid et al., 2022).

1.4 Research Direction and

Contribution

The key challenge lies in the effective utilization of

the collected data to drive improvements in the

operations management of outpatient chemotherapy,

rather than just digitizing the process and analyzing

the data. The ability to translate available data into

concrete actions is what sets Data-Driven Operations

Management Research (DDOMR) apart from other

data-driven research, such as data-driven research of

economists and statisticians or empirical Operations

Management (OM) research (Simchi-Levi, 2013).

Therefore, DDOMR has the potential to bridge the

gap between prior studies and real-world operations

by using data to identify areas for improvement and

provide novel solutions (Gupta et al., 2021). This

paper builds on this research direction and uses

DDOMR to address gaps in the management of

outpatient chemotherapy operations.

Through the collection and analysis of extensive

data, this research identifies several avenues for

further exploration, with a focus on using Machine

Learning (ML) models to predict key outcomes and

improve operations management. This study

contributes to the development of new ML models

that support decision-making and optimization in

outpatient chemotherapy. The models developed and

tested in this study show promising results in

predicting length of stay, admission delay, changes in

patient weight, and drug administration duration.

Furthermore, the study highlights the importance of

patient, drug, and process features in predicting these

outcomes. This contribution provides a foundation for

further research in this field and a roadmap for future

improvements in the operations management of

outpatient chemotherapy.

2 LITERATURE REVIEW

The literature on OCOM focuses on several key

themes, primarily the optimization of patient flow

through better appointment scheduling and planning,

coordination between departments in outpatient

chemotherapy centers, and resource allocation.

Operations management scholars have approached

these issues using various optimization and

simulation models to improve a wide range of

performance indicators (Lamé et al., 2016). The

models used in the literature to optimize outpatient

chemotherapy can be classified into deterministic,

stochastic, and data-driven (Hadid et al., 2021).

The dearth of research in OCOM using cutting-

edge data analytics such as ML models highlights a

significant gap. Currently, no studies have combined

ML models with decision support models in OCOM.

Although, two studies (Mosa et al., 2021; Smith &

Carlson, 2021) are relevant to the topic: the former

predicts the risk of chemotherapy-induced nausea and

vomiting using ML, while the latter suggests an ML-

based approach to decrease emergency admissions

due to chemotherapy side effects. This limited

number of articles underscores the need for further

data collection and utilization in OCOM research.

ML has been widely used in healthcare operations

management, particularly in the field of cancer

services. Numerous studies have shown its potential to

improve the efficiency and quality of care provided to

patients (Pianykh et al., 2020). For instance, ML

algorithms have been used to predict the demand for

certain cancer services, enabling healthcare organiza-

tions to manage their resources more effectively

(Bastani et al., 2020) and providing more accurate and

personalized care plans (Almeida & Tavares, 2020).

Nevertheless, despite its promising results, the

integration of ML into healthcare operations

management is still in its early stages and further

Leveraging Health Informatics to Enhance Outpatient Chemotherapy Operations Management

379

research is needed to fully realize its potential. The

search results from the Scopus database using the

query in Table 1 indicate a gap in the use of ML

models to improve the Outpatient Chemotherapy

Process (OCP). Out of the 92 articles found, none

used ML models to predict key factors affecting the

OCP management. This further highlights the need

for further research to utilize ML to improve the OCP

(Hadid et al., 2021, 2022).

Table 1: The Search Query Used to Collect Articles from

Scopus Database.

TITLE-ABS-KEY ( "machine learn*" OR "artificial

intelligence" OR "pattern* recognition*" OR "feature*

selection*" OR "deep neural network" OR "deep

learning" OR "convolutional neural network" OR

"artificial neural network" )

AND TITLE ( chemotherapy )

AND TITLE-ABS-KEY ( predict* OR forecast* )

AND NOT TITLE-ABS-KEY ( response )

This study aims to contribute to this research

stream by developing new ML models that support

decision-making and optimization in outpatient

chemotherapy. The focus is on predicting patient

behavior and key outcomes related to the

administration of chemotherapy drugs, such as

appointment delays, drug administration durations,

and drug dose variation.

3 PROBLEM DESCRIPTION

We have conducted a time series analysis of

operational data to uncover patterns and potential

issues in daily activities. Figure 1 shows a significant

difference between planned and actual patient

admissions. The center allows early admissions to opti-

mize bed utilization and hasten drug order activation

and preparation processes. Patients are predominantly

admitted at two time slots, 7 AM and 11 AM, with

more patients arriving early for the 11 AM slot.

However, as depicted in Figure 2, the number of

patients receiving the primary service of drug

administration is significantly smaller compared to

those waiting or receiving secondary services. During

peak hours, the proportion of patients receiving drugs

does not surpass 15% of the total number of patients

in the center, which contradicts the expectations from

allowing early admissions and punctual patient

arrivals to reduce crowding and improve patient flow.

These factors make predicting the length of stay

difficult, as shown in Figure 3. There are multiple

medical and operational factors that contribute to the

Figure 1: Average daily planned and actual admission

patterns for one month.

Figure 2: Average daily patterns of number of patients in

the center and patients administering drugs for one month.

length of stay. The goal is not to study these factors,

but to highlight the challenge in accurately predicting

the length of stay, which is used as an input in

appointment planning and scheduling. The drugs net

infusion duration is not a representative of the length

of stay, even if the expected time for a nurse to change

the drug between drug infusions is added to it to

calculate the total drug administration duration.

Despite this, the total drug administration duration is

still not a reliable predictor of the length of stay as

shown in Figure 4.

Figure 3: Average daily percentage of number of patients in

main appointment stages over time for one month.

Figure 4: Comparison between length of stay, drug

administration duration, and net drug infusion duration for

737 appointments.

HEALTHINF 2024 - 17th International Conference on Health Informatics

380

Therefore, there is a need for predictive models to

enhance decision support in outpatient chemotherapy.

Our analysis highlights the importance of predicting

four key factors: (1) change in patient body weight

between dosing day and appointment day, (2) delay

from the planned admission time, (3) drug

administration duration, and (4) length of stay. These

predictions are crucial for all departments involved,

particularly pharmacy and day care, as they can assist

in improving the accuracy of drug preparation and

scheduling. The prediction of changes in weight and

expected admission delays can help the pharmacy

increase the number of drugs prepared in advance.

Predictive models for drug administration duration

and length of stay can improve scheduling efficiency

and increase capacity utilization. Consequently, this

will reduce the gap between the percentage of drug

orders dispensed and the percentage of drug orders

initiated for administration, ultimately reducing the

number of patients present in the clinic during the day.

4 METHODOLOGY

As discussed in the previous section, we aim to

demonstrate the potential of using ML models to

support decision making in outpatient chemotherapy

centers by predicting four important outcomes: (1)

change in patient body weight, (2) length of stay, (3)

admission delay, and (4) total drug administration

duration. The following subsections presents the data

preparation and model development steps.

4.1 Feature Selection

We identified the following features for our ML

models: age, gender, body weight, oral body

temperature, height, respiratory rate, blood pressure,

type of drugs used (9 categories), drug administration

duration for all cycles (1 to n-1), cancer type

(oncology or hematology), number of drugs used,

appointment slot (morning or afternoon), prior notice

for appointment in days, and day of appointment

(Sunday to Thursday). The type of drugs was encoded

into 9 categories, and days of appointment were

encoded into 5 categories, resulting in a total of 26

distinct features for predicting the desired outcomes.

4.2 Dataset

The dataset used for each of the four outcomes

consisted of 1121 rows of encounter details, 26

features, and the target outcome. However, due to the

presence of 817 missing values in the body weight

feature, only 304 encounter details were used to build

the predictive model for body weight variation. For

the other three outcomes, all 1121 data points were

utilized.

4.3 Data Preprocessing and Model

Building

For features with less than 20% missing values, KNN

imputation was used to estimate the missing values.

To predict all four outcomes, we developed linear

regression and Xgboost regression models by

splitting the available data into 80-20 train-test splits.

The predictive models were developed using the

scikit-learn library in Python, and the mean absolute

error metric was used to evaluate their performance

on the 20% test data set.

5 RESULTS AND DISCUSSION

5.1 Predication of the Length of Stay

The length of stay of a patient is a crucial factor in the

management of hospital resources and bed capacity.

The traditional approach of estimating length of stay

by adding a fixed time to the drug infusion duration

is often insufficient in accurately predicting the actual

length of stay. This can lead to inefficiencies in bed

management and potentially result in overbooking or

underutilization of resources.

To address this issue, a ML model was developed

in this study to predict length of stay more accurately.

The model was trained on patient and process data,

including features such as age, height, blood pressure,

days to the appointment, number of drugs,

temperature, respiratory rate, cancer type (tumor or

blood cancer), drug infusion durations, day of

appointment, and type of drug. Results from the study

showed that these features are relevant for predicting

length of stay (Figure 5).

The utilization of this ML model will allow

schedulers to make more informed decisions and

accurately estimate length of stay, leading to a more

efficient management of bed capacityPredication of

the body weight variation

The process of determining the appropriate dosage of

drugs for patients is a critical aspect in oncology

treatment. One of the factors that affects the dose is

the change in the patient's body weight. In order to

calculate the percentage change in body weight, the

difference between the dosing weight and the

measured weight on the appointment day is divided

by the dosing weight and multiplied by 100.

Leveraging Health Informatics to Enhance Outpatient Chemotherapy Operations Management

381

Figure 5: Actual and predicted values of length of stay using

the two models as well as the top 15 relevant features to

predict the outcome.

Figure 6: Actual and predicted values of body weight

variation using the two models as well as the 10 most

relevant features with respect to the outcome.

The dosing day, which is usually before the

appointment day, is when the patient comes to the

center to be seen by the oncologist and have his

weight measured and drug order placed. However, the

patient's weight may change before the appointment

day due to various reasons. As a result, the nurse

checks the weight on the appointment day to confirm

that it is still within 5% error and the same drug order

is still valid.

The pharmacy does not prepare the drugs before

the weight is measured due to the possibility of a

weight change greater than 5% and the subsequent

waste of the prepared drug. Therefore, to predict the

change in body weight more accurately, the XGboost

model was used, which showed better and more

accurate predictions compared to the linear regression

model (Figure 6). The most important features for

prediction were found to be age, vital signs, and the

number of days between the dosing and appointment

days.

Figure 7: Actual and predicted values of delay between

planned and actual admission using the two models as well

as the top 15 relevant features to predict the outcome.

Figure 8: Actual and predicted values of drug

administration duration using the two models as well as the

top 15 relevant features to predict the outcome.

5.2 Predication of Admission Delay

The results of predicting admission delay are

presented in Figure 7. The performance of two

models, linear regression and XGboost, is evaluated

on a test dataset. The results highlight the importance

of considering different factors that may influence

admission delay. One of these factors is the

appointment slot, which can either be 7 am or 11 am.

This factor is found to be relevant to admission delay

and can be used to enhance the overall patient

experience by improving the scheduling system.

Another important factor is the number of days

between the date when the patient was notified about

the appointment and the actual appointment date.

This factor can provide insights into the patient's

behavior and how early or late they are notified about

the appointment. These results highlight the

significance of studying patient behavior based on the

appointment time and notification date, which can

provide valuable information for optimizing the

scheduling system.

 

 

  

HEALTHINF 2024 - 17th International Conference on Health Informatics

382

5.3 Predication of Drug Administration

Duration

In an outpatient chemotherapy center, beds and

infusion chairs are critical resources. Predicting the

length of stay of patients in these resources is crucial

for efficient resource management. Although the net

duration of drug infusion is known and fixed, patients

may require additional time during administration

due to the preparation and removal of the drip,

monitoring of their health status, and post-infusion

observation.

Figure 8 highlights that the type of drugs being

administered, such as monoclonal antibodies,

antimetabolites, and alkylating agents, are among the

top 15 relevant features that impact the drug

administration duration. This highlights the

importance of considering these factors when

predicting the drug administration duration in the

beds and infusion chairs.

5.4 Comparison of Models

In this study, the results demonstrate the potential of

using ML models for improved decision-making in

outpatient chemotherapy centers. As shown in Table

2, the Xgboost model performed better than linear

regression in terms of mean absolute error (MAE),

suggesting its effectiveness in predicting outcomes.

However, it is worth noting that a simple linear

regression model was also able to provide meaningful

predictions, indicating that the suggested operational

characteristics (OC) features and ML models can

provide valuable insights.

Table 2: Performance of the ML models for four outcomes

for outpatient chemotherapy operations management.

Mean Absolute error

Length

of stay

Body

weight

variation

Delay

Infusion

duration

Linear

regression

205.17

4.99

0.46

0.48

Xgboost

134.23

3.95

0.43

0.32

Despite the positive results, it is important to

acknowledge the limitations of this study. One such

limitation is the classification of drugs into 9

categories, which was based on mechanism of action

and compound type. However, many drugs have

overlapping features and were not accounted for in

the classification process. For example, drugs like

ado-trastuzumab emtansine and brentuximab vedotin

have properties of both monoclonal antibodies and

cytotoxic drugs, but were classified under

monoclonal antibodies for simplicity. Additionally,

the number of drugs used was accounted for in the

feature "number of drugs", but the frequency of use

of each type of drug was not considered. Furthermore,

the ML models used in this study may benefit from

further fine-tuning of hyperparameters. Addressing

these limitations has the potential to further enhance

the overall performance of the models.

6 CONCLUSION

In conclusion, the present study aims to address the

gap in data-driven research in outpatient

chemotherapy (OCOM) operations management.

Extensive data was collected, and research

opportunities were identified. In particular, the use of

ML models, was thoroughly explored and tested. The

results show that XGboost model outperformed linear

regression in terms of mean absolute error in

predicting admission delay and drug administration

duration. These findings demonstrate the potential of

using ML models in OCOM to improve decision-

making, resource allocation, and patient satisfaction.

However, the study also highlights some limitations

and avenues for future improvement. Further research

is needed to address these limitations and fully

harness the potential of data analytics and ML in

OCOM field.

ACKNOWLEDGEMENTS

This article was made possible by National Priorities

Research Program -Standard (NPRP-S) Twelfth

(12th) Cycle grant# NPRP12S-0219- 190108, from

the Qatar National Research Fund (a member of Qatar

Foundation). The findings herein reflect the work,

and are solely the responsibility, of the author[s].

REFERENCES

Ahmadi-Javid, A., Jalali, Z., & Klassen, K. J. (2017).

Outpatient appointment systems in healthcare: A

review of optimization studies. In European Journal of

Operational Research (Vol. 258, Issue 1).

https://doi.org/10.1016/j.ejor.2016.06.064

Almeida, G., & Tavares, J. M. R. S. (2020). Deep Learning

in Radiation Oncology Treatment Planning for Prostate

Cancer: A Systematic Review. In Journal of Medical

Systems (Vol. 44, Issue 10). https://doi.org/10.1007/

s10916-020-01641-3

Leveraging Health Informatics to Enhance Outpatient Chemotherapy Operations Management

383

Bastani, H., Zhang, D., & Zhang, H. (2020). Applied

Machine Learning in Operations Management. SSRN

Electronic Journal. https://doi.org/10.2139/ssrn.3736

466

Evans, J. M., Qiu, M., MacKinnon, M., Green, E., Peterson,

K., & Kaizer, L. (2016). A multi-method review of

home-based chemotherapy. In European journal of

cancer care (Vol. 25, Issue 5). https://doi.org/10.1111/

ecc.12408

Gupta, S., Starr, M. K., Farahani, R. Z., & Asgari, N.

(2021). OM Forum—Pandemics/Epidemics: Challenges

and Opportunities for Operations Management Research.

https://doi.org/10.1287/msom.2021.0965, 24(1), 1–23.

https://doi.org/10.1287/MSOM.2021.0965

Hadid, M., Elomri, A., El Mekkawy, T., Jouini, O.,

Kerbache, L., & Hamad, A. (2021). Operations

Management of Outpatient Chemotherapy Process: An

Optimization-Oriented Comprehensive Review.

Operations Research Perspectives, 100214.

Hadid, M., Elomri, A., El Mekkawy, T., Kerbache, L., El

Omri, A., El Omri, H., Taha, R. Y., Hamad, A. A., &

Al Thani, M. H. J. (2022). Bibliometric analysis of

cancer care operations management: current status,

developments, and future directions. Health Care

Management Science, 1–20.

Houts, P. S., Lipton, A., Harvey, H. A., Martin, B.,

Simmonds, M. A., Dixon, R. H., Longo, S., Andrews,

T., Gordon, R. A., Meloy, J., & Hoffman, S. L. (1984).

Nonmedical costs to patients and their families

associated with outpatient chemotherapy. Cancer,

53(11). https://doi.org/10.1002/1097-0142(19840601)

53:11<2388::AID-CNCR2820531103>3.0.CO;2-A

Lamé, G., Jouini, O., & Stal-Le Cardinal, J. (2016).

Outpatient chemotherapy planning: A literature review

with insights from a case study. IIE Transactions on

Healthcare Systems Engineering, 6(3), 127–139.

https://doi.org/10.1080/19488300.2016.1189469

Mandelbaum, A., Momčilović, P., Trichakis, N., Kadish,

S., Leib, R., & Bunnell, C. A. (2019). Data-Driven

Appointment-Scheduling Under Uncertainty: The Case

of an Infusion Unit in a Cancer Center.

Https://Doi.Org/10.1287/Mnsc.2018.3218, 66(1), 243–

270. https://doi.org/10.1287/MNSC.2018.3218

Mosa, A. S. M., Rana, M. K. Z., Islam, H., Mosharraf

Hossain, A. K. M., & Yoo, I. (2021). A smartphone-

based decision support tool for predicting patients at

risk of chemotherapy-induced nausea and vomiting:

Retrospective study on app development using decision

tree induction. JMIR MHealth and UHealth, 9(12).

https://doi.org/10.2196/27024

Pianykh, O. S., Guitron, S., Parke, D., Zhang, C.,

Pandharipande, P., Brink, J., & Rosenthal, D. (2020).

Improving healthcare operations management with

machine learning. Nature Machine Intelligence, 2(5).

https://doi.org/10.1038/s42256-020-0176-3

Simchi-Levi, D. (2013). OM Forum—OM Research: From

Problem-Driven to Data-Driven Research.

https://doi.org/10.1287/msom.2013.0471, 16(1), 2–10.

https://doi.org/10.1287/MSOM.2013.0471

Smith, M., & Carlson, J. (2021). Reducing ED Visits and

Hospital Admissions after Chemotherapy with

Predictive Modeling of Risk Factors. Oncology Issues,

36(4). https://doi.org/10.1080/10463356.2021.1927638

Waller, A., Forshaw, K., Bryant, J., & Mair, S. (2014).

Interventions for preparing patients for chemotherapy

and radiotherapy: A systematic review. In Supportive

Care in Cancer (Vol. 22, Issue 8).

https://doi.org/10.1007/s00520-014-2303-3

HEALTHINF 2024 - 17th International Conference on Health Informatics

384