Combating sequential time delays in pharmaceutical supply chains

By Anirudh Narula and Yu-Hsin Lin

Editor’s Note: The SCM capstone Recurrent Neural Network for Predicting Sequential Supply Chain Delays was authored by Anirudh Narula and Yu-Hsin Lin and supervised by Tim Russell ([email protected]). For more information on the research, please the thesis supervisor.

Ahead of the clock: Leveraging machine learning to provide early warning signals

The pharmaceutical industry faces significantly longer lead times than other industries, and time delays can seriously impact the health outcomes of countless individuals. Recognizing the potential risks this poses, GlaxoSmithKline (GSK) aims to improve on-time delivery through an end-to-end early warning system.

GSK’s global supply chain is intricate, involving both primary (active pharmaceutical ingredients manufacturing) and secondary (finished product creation) locations. Therefore, the first challenge was tracking planned dates to identify where and when delays occur. The second challenge was managing the domino effect of these sequential delays, which can jeopardize timely medication delivery to patients.

Machine learning is not enough

To tackle this issue, we developed an early warning system designed to preempt delays. Our goal was to build a machine-learning model capable of predicting the cascading impacts of delays on specific batches or shipments. Before delving into modeling, we analyzed summary statistics to understand GSK’s supply chain dynamics. By examining metrics, such as mean and standard deviation of delays, and the distribution of delay occurrences, the analysis identified potential points of improvement and pinpointed areas of risk within GSK’s supply chain.

Next, we established planned dates for reliable tracking within GSK’s system. We reviewed historical data on production, quality, and delay frequency. This analysis helped identify the standard performance benchmarks. By integrating these data points, we could effectively track actual progress against the planned dates as the deviations signal potential delays.

We built a recurrent neural network (RNN) machine learning model to provide early warning signals of sequential time delays. The RNN model suits GSK’s supply chain due to its ability to process sequential data and recognize patterns over time. To better interpret the model results, we also developed an explanatory model with Shapley Additive Explanations (SHAP) values. Higher SHAP values indicated features with a greater impact on predictions, which helped identify key drivers causing time delays.

After hyperparameter tuning, the RNN model achieved a mean absolute error (MAE) of 4.89 days. This precision suggests it is effective in predicting delays. Additionally, the majority of prediction errors are clustered around 0, indicating highly accurate predictions. The SHAP values from the explanatory model revealed that specific path sequences within GSK significantly affect predictions. With deeper analysis across all supply chain stages, we further pinpointed that the assembly stage has the most positive impact on time delays. Based on this finding, GSK inferred that the quality stage preceding assembly could be the primary driver of delay impacts.

Now, what’s next?

To extend the model to other brands or sites in the future, we recommended that GSK start with summary statistics to identify key delay factors before jumping into RNN modeling. Integrating this analysis into the existing GSK front-end dashboard can transform it into an early warning system, providing a holistic view of delay issues for proactive resource allocation. To further enhance the value of the RNN model, we developed another prediction model for forecasting new data. This framework empowers GSK to precisely pinpoint bottlenecks and promptly alert relevant departments to take action.

This project has paved the way for future advancements in GSK’s supply chain management. By addressing the scarcity of prior studies on sequential delay issues in pharmaceutical supply chains, this project also contributes valuable insights to the existing research domain. The model and findings can reduce delays and potentially lower safety stock inventory. Building on this foundation, GSK can enhance efficiency, mitigate supply chain risks, and deliver essential medications to patients more effectively.

Every year, approximately 80 students in the MIT Center for Transportation & Logistics’s (MIT CTL) Master of Supply Chain Management (SCM) program complete approximately 45 one-year capstone projects.

These students are early-career business professionals from multiple countries, with two to 10 years of experience in the industry. Most of the research projects are chosen, sponsored by, and carried out in collaboration with multinational corporations. Joint teams that include MIT SCM students and MIT CTL faculty work on real-world problems. In this series, they summarize a selection of the latest SCM research.

SC
MR