Data entry

Entering data safely

Decorative image of fingers using a keyboard

Entering data correctly, whether numbers or text, is a routine part of medical treatment. Examples include entering personal details into patient records, setting the amount of a drug that an infusion pump should deliver to a patient, recording prescriptions, and so on. Unfortunately, there is plenty of scope for mistakes to be made that could ultimately harm patients. Some mistakes involve the wrong information being entered. Others result in the wrong sequence of steps being followed. We have explored the underlying reasons why these kinds of mistake happen and how they can be avoided. We have looked at both the limitations of the way our brains work (i.e. our cognition), and how changes to the design of data entry systems can help prevent such errors leading to harm.

Interruptions and short term memory

Interruptions tax people’s short term memory. This makes errors more likely to occur. However, our experiments have shown that if you give yourself a few extra moments to recall what you were doing before you are interrupted, your memory is more accurate. Well-designed visual reminders support your short term memory and make it easier to pick up where you left off and finish the task correctly.

Entering numbers safely

A particular focus of the CHI+MED data entry project has been to focus on number entry. Entering numbers is typically a part of operating computers and computer-based gadgets, including medical devices where safety is an issue. The purpose of any number entry interface is simple – to select a specific value. The interfaces used to do this can however be deceptively complicated as we have shown in devices ranging from microwave ovens to infusion pumps. Worse, the way these interfaces are implemented ignores user error and can result in unpredictable and is potentially unsafe, so could lead to patients being harmed.
We have aimed to discover and understand the types of errors common in number entry tasks, explore what design factors influence these errors and build fundamental understanding of how people think about number entry and what effect an interface has on their thinking.

Key points

  • We have compared different interfaces for entering numbers, determining the safest.
  • We have shown that by prompting people they are less likely to make mistakes.
  • We have shown that if some numbers are more familiar than others and this affects the way numbers are entered, and experiments on number entry must be designed accordingly.
  • We have also developed mathematically based tools that help find problems in the designs of number entry systems.

Designing the safest number entry interfaces

We have evaluated existing interfaces for entering numbers and shown that the style of interface influences the type of error committed and its severity. With existing designs there is a trade-off between speed and safety. Digit-based keypads are fast but likely to lead to errors that are out by 10 or more, whereas 5-key interfaces are slower but less likely to lead to large and dangerous mistakes. When keying errors are made, like trying to increase the number beyond the largest possible, they should be blocked with a warning that should be acknowledged. Entering numbers can be made easier if the interface is tailored based on which numbers are most commonly used for the intended task.

Using priming questions to reduce entry errors

We have also shown that prompting people with questions linked to the number they are about to enter makes it more likely that they will enter it correctly. It can also help them correct mistakes when they do make them. In particular, people made far fewer errors after being asked questions about the format of the number, the quantity it refers to, and the context of the number entry task.

Familiar numbers

There are patterns in the numbers that medical workers have to type into devices used on hospital wards. This suggests that some numbers could be more familiar than others. The familiarity of a number has a significant effect upon how a user types it: familiar numbers are faster to type than non-familiar numbers. Often number entry systems are tested by asking participants to enter sets of random numbers, but this doesn’t accurately reflect the real tasks performed in hospitals. We have shown that future experiments on number entry need to be designed in a way that takes account of whether numbers used are familiar to the participants.

A Hazard Analysis for Infusion Pump User Interface Software

Working with the US regulator, the Food and Drug Administration (FDA), we have: identified a substantial set of use-related hazards for infusion pumps data entry systems, and explored systematic analysis techniques for identifying use-related hazards.

Developing tools for safer number entry

Developing easy to use interfaces for entering data, and especially numbers, is surprisingly hard. As part of CHI+MED we explored what makes a good number entry interface, comparing different kinds, for example, to see which led to people making most mistakes. We have also developed tools to help designers create good data entry interfaces in the first place, and also to help them evaluate the interfaces they create.

Programming safer keyed data input

We have created a tool that automatically creates data entry user interfaces that are safer to use than those created in ad hoc ways. It helps ensure that programmers cover all eventualities and the user interface responds to errors in a sensible way. User interfaces cannot be made perfectly safe. This is a fundamental problem we define and call divergence. It means that whatever is done to prevent users making mistakes there will always be some other mistake that can still be made. Programmers have at best ad hoc solutions over what to do when people enter data incorrectly, like when they include a second decimal point or put a leading 0 in a number, but often programmers just ignore these seemingly simple problems. Unfortunately, failing to manage user error well causes even worse problems for users, and this is what our tool helps prevent. One approach we developed that the tool uses is traffic lights, so the user interface and the user know when a mistake has been made. A red traffic light – perhaps augmented with a sound or vibration (and it could be any colour, or flash for colour-blind users) – alerts the user that a problem must be solved. With the tool, traffic lights are implemented flexibly, efficiently and dependably. Crucially, the programmer can then build software that does not need to handle the user errors, since the tool has done that already.

Stochastic evaluation: randomly simulating users pressing buttons

Many different user interfaces for entering numbers – as found on devices from office products to medical devices – could be made much safer without affecting their normal use. Our technique, called stochastic evaluation, which involves computer simulation of users pressing buttons, is a very fast and effective way of seeing how usable and safe real or proposed designs are. We have identified ways to make common designs much safer, quickly identifying bugs or defective design choices. Our approach also allows choices to be compared by putting a value on the size of a problem’s impact on safety. The technique is automatic and best when there is an objective measure of how significant an error is. For example, ‘drug doses that are out by a factor of ten or more’ is easy to quantify and hence easy to evaluate automatically.

Using control theory to model interaction with medical devices

We have developed a new approach for evaluating the user interfaces of medical devices based on techniques from aerospace engineering and robotics. It combines manual control theory with hybrid automata. We have modelled both discrete and continuous human operator behaviour and, in particular, we have found a novel way to model a person using a medical device with up and down buttons. We can model both short button presses and holding down buttons. We have integrated the model with the CHI+MED analysis toolset, PVSio-web, and have demonstrated the model’s utility by simulating a person entering values on an infusion pump, using up and down buttons. A typical insight is the identification of user interface design flaws that would lead to a large overshoot when using up and down buttons to adjust numbers. It also helps an analyst to identify and reflect on different types of user behaviour.