ZAN Project Overview
Zoey's Motivation
Working with real world data provided to us by Hospital X is interesting. People’s behaviour, even when they follow expectations, are often varied and unique, and there is always a possibility of observing unexpected behaviours in the data. While we start off seeking improve the productivity of our sponsor’s colleagues, there is no telling what our analysis of the data might uncover about the situation and the problem, and that makes the project exciting.
It is very encouraging that our work has the potential to impact people’s lives. The sponsor’s concern for the community he works in is motivating as well. That has led him to do more for the hospital, even on top of his normal duties, inspires me to want to dig through the data and see if we can find anything that could him help the hospital improve their processes.
Aishwarya's Motivation
Many hospitals and clinics often face the problem of cancelled appointments as well as no show by patients. The time allocated by the doctors and consultations for these patients goes to waste, time that could have been significantly used by another patient in need of it. Our motivation is to be able to come up with a solution to efficiently maximize the utilization of the doctor’s time to be able to serve as many patients as possible. In this process, the doctors do not have to spend their time allocated to patients idly.
Another possible problem that we wish to look into through this project is the link between a patient’s details and his/her probability of cancelling an appointment. There could be various factors that influence the chance of a patient not showing up, such as the time of the appointment, the location of the patient, the age of the patient, experience of previous visit and so on. We hope to study these factors and try to find a possible link between the two. In doing so, we would be able to predict the probability of a patient not turning up for the appointment, and subsequently allocate the time to another patient in need.
Nas's Motivation
This project offers a follow-up to a previous project that I have done with a subsidiary of Hospital X. During that project, I have learn a lot about the organization and the amount of effort put in by the various stakeholders in order to improve the mental wellbeing of the general population. As an operationa management student, I am keen in exploring ways in improving the productivity of the staff as well as maximizing the access of care to the patients.
This project interests me as it offers a unique opportunity to further explore an unfamiliar domain (the medical sector). I believe that we can learn much from our project sponsor as he is a champion for data analytics and has considerable experience.
The objectives of the project would be the following:
- Business objective: To identify factors that relates to no-show appointments and predicts patients’ attendance rate in order to improve Hospital X’s scheduling of appointments and utilisation of appointment slots.
- Technical objective: To use data analytical tools and statistical methods to study the data and obtain insights that would facilitate the business objective.
- To understand the data domains
- To understand the workflow of scheduling a patient’s consultation process
- To identify the contributing factors that lead patients to defaulting appointments
- To conduct what-if analyses to understand changes in appointment rates if the patient is referred to a medical professional nearer to them
- To evaluate the feasibility of creating a predictive model
The dataset is based on Hospital X's child and adolescent department inpatient records. The inpatient records are processed by the hospital staff working on the front desk. The patient visits are mainly categorised into 1) first appointment with a doctor, 2) review appointment with a doctor, 3) first appointment with a psychologist and 4) reviewed appointment with a psychologist.
For this project, we will follow Data Analytics Lifecycle approach closely.
With reference from several research papers such as Michelle. K. (2011). and Molfenter. T. (2013), our secondary findings would be the following:
- Younger patients are significantly less likely to keep their initial outpatient mental health appointments
- No-show behavior is positively correlated with lower income and lower socioeconomic status
- Previous appointment experience of the patient, such as number of previous appointments, their types and lead times, do play a part in a patient defaulting his or her appointment
- The longer a patient has to wait for an appointment to be scheduled, the less likely is the patient to keep his or her first appointment
These secondary findings will be useful as a starting platform for us to carry out the analysis. We will test the given data against the secondary findings to see if there is any conformity.
While the project will revolve around the above objectives, our project sponsor is flexible to allow us to explore other possible relevant analytical tools or techniques that would enhance the findings.
- The dataset is limited to records from 2015 to 2016, which prevent any seasonal or yearly analysis
- The dataset only pertains to our project sponsor’s department, which administers only younger patients of ages 18 years and below.
Phase 0: Learning about the Case Context
We will gather and map out all information on patient appointment scheduling. This includes:
- Mapping out the workflow in Hospital X’s context, as well as the general process as described in existing literature
- Consolidating significant factors of no-show from literature review
- Review existing methods/recommendations for the problem of no-show
Phase 1: Data Cleaning
In the first phase, we will closely study the dataset to understand each of its variables and values so as to prepare it for analysis. This involves the following steps:
- Recording the description and range for each variable and its values
- Identifying irrelevant or duplicate fields
- Resolving missing and invalid values
- Cross-check related variables to verify accuracy
- Transform skewed variables, merge variables for ease of analysis
- Record assumptions made
- Documenting all of the above
Depending on our findings regarding the dataset and its errors, we will also consult with the sponsor to understand more specifically on questions of how some fields and its items are entered.
Phase 2: Data Exploration
In the second phase, we will conduct exploratory data analysis, as well as test some of the hypotheses gleaned from our literature review findings.
Exploratory data analysis steps include:
- Studying the distributions of variables
- Identifying and treating outliers/anomalies
- Finding clusters or groupings (cluster analysis)
- Compare two or more locations or time periods (any cycles/ seasonal trends)
- Examine relationships between variables (regression analysis)
- Develop hypotheses based on literature and conduct hypothesis testing
This analysis should go through a number of iterations, as we will continually compare our findings to existing literature as well as what we know of Hospital X’s processes.
Phase 3: Data Modelling
By the final phase, we would have a good understanding of the data and case, and develop models for predicting no-show in patients. From this, we will be able to develop solutions for Hospital X.
Steps include:
- Develop various (multiple regression) models
- Compare models and select best model based on testing
- Interpret model to develop strategies that Hospital X can adopt
