Clinical Outcome Assessments (COA) are designed to give structure to the patient voice. By asking people to report on their symptoms, functioning and wellbeing, they provide critical data for trials, treatment evaluation and regulatory decision-making. Yet COAs depend heavily on something fragile: the patient’s ability to remember, evaluate and report experiences over time.

That ability is constrained by the limits of working memory. This is the short-term system that holds and manipulates information. Cognitive psychology has shown that working memory is strictly limited in capacity, closer to four “chunks” than seven-plus-or-minus-two once suggested. But capacity is only part of the story. Working memory is also subject to biases: systematic distortions in how information is recalled, aggregated and expressed.

These biases mean that COA responses are not neutral records of what happened. Instead, they are filtered approximations, shaped by what is easiest to remember or the most salient in the moment. This article explores the core working memory biases most relevant to COA responding: recency bias, salience, mood-congruent recall, averaging, interferences, anchoring to extremes and omission.

Recency Bias

One of the most reliable features of human memory is recency bias. This is the tendency to give more weight to the most recent events.

In a COA that asks about symptoms “over the past month”, patients are more likely to recall what happened in the last few days than what happened three weeks ago. If the recent period was unusually good or unusually bad, it can dominate the overall rating.

For example, a patient who was fatigued throughout most of the month but felt better in the last week may underreport fatigue. Another who was stable until the final days of a flare-up may overreport this.

Impact: Recency bias reduces accuracy of long recall windows. Responses are skewed towards the end of the period.

Salience and Vividness Bias

Memory does not treat all events equally. Vivid, emotionally charged or unusual experiences are remembered more readily than routine ones. This is salience bias, also called the vividness effect.

In COAs, this means that one striking episode can outweigh weeks of ordinary experience. A single night of agonising pain, a vivid nightmare or a severe panic attack may dominate recall, even if it happened only once in the timeframe.

The patient then endorses a higher frequency or severity category than their average experience warrants.

Impact: Responses reflect the emotional salience of events rather than their true frequency, skewing data towards the extreme.

Mood-Congruent Recall

Our current mood colours what we remember. When feeling low, negative consequences are more accessible; when feeling positive, positive experiences come more easily. This is mood-congruent recall.

In COAs, a patient’s mood at the moment of answering influences when they recall and how they rate it. A patient with depression filling out a questionnaire on a “bad day” may recall more episodes of hopelessness and rate symptoms as more frequent. The same patient on a “good day” may recall fewer episodes and report less severity.

Impact: Mood-congruent recall reduces test-retest reliability. Responses vary with the state of the patient at the time of completion, rather than reflecting a stable summary of experience.

Averaging Bias

When asked to summarise experiences across time, people rarely calculate precise frequencies. Instead, they take shortcuts, compressing complex histories into a simple average of “gist”. This is averaging bias.

For example, a patient who had 20 good days and 10 bad days in a month may not distinguish between the two. They may simply report “Sometimes” for a symptom, even though the lived pattern was more like “rarely” for most of the month, punctuated by concentrated clusters of severity.

Impact: Averaging bias masks variability and smooths out fluctuations. Instruments may fail to detect change over time if patients report rough averages rather than precise counts.

Interference and Overlap

Working memory struggles when items are similar. Experiences that share features can interfere with each other, becoming blurred or confused.

In COAs, this manifests when descriptors overlap. For example, pain described as ‘stabbing’ versus ‘cutting’, or emotional effects like “loss of motivation” versus “depression”. Patients may be unsure whether to endorse both, one, or neither.

Interference also arises when multiple symptoms occur together, such as fatigue, pain and low mood, making attribution to a single construct difficult.

Impact: Overlap leads to inconsistent responding, double-counting or omission. Reliability suffers when items are semantically close.

Anchoring to Extremes

When recalling experiences over time, people often anchor their judgement to the most extreme episodes. A single very bad day overshadow weeks of moderate days.

For example, a patient who had one severe panic attack in a month may report “Often” because that day looms large in memory. Conversely, a single period of relief may lead to underreporting.

Impact: Anchoring skews responses towards extremes, inflating or deflating scores depending on whether salient outliers were positive or negative.

Capacity-Driven Omission

Working memory cannot hold all relevant events in mind at once. As patients recall, less salient or routine events are dropped. This is omissions bias, driven by capacity limits.

Mild or moderate daily symptoms may be forgotten, while severe but rare episodes are retained. A patient with frequent but unremarkable joint stiffness may underreport, while remember only flare-ups.

Impact: Routine, low-intensity symptoms are systematically underrepresented in COA data.

Pulling It Together

These biases rarely happen in isolation. A patient completing a COA may be tired (lowering capacity), in a low mood (increasing negative recall) and influenced b a recent vivid episode (salience and recency). The combination means their answer reflects a biased sample of memory, not the whole period.

What looks like a reliable, numeric response on a scale is actually the product of cognitive shortcuts. That does not make the data useless, but it does mean that it must be interpreted as caution.

Implications for COA Design

Recognising these core working memory biases suggests way to reduce distortion:

• Use shorter recall period, such as one week rather than one month, to reduce recency and omission biases.

• Provide clear response anchors with examples (“2-3 times per week”) to counter averaging and anchoring biases.

• Do not use overlapping descriptors; test translations carefully to reduce interference.

• Consider momentary electronic capture for symptoms prone to vividness bias, so severe episodes do not dominate memory.

• Be aware of mood effects: a single administration captures a state-filtered perspective, not a neutral record. Multiple administrations may give a more balanced view.

Conclusion

Clinical Outcome Assessments depend on memory, but memory is biased and flawed. Recency, salience, mood, averaging, interference, anchoring and omission all shape how patients respond. The result is that COAs capture a filtered and compressed version of lived experience.

By acknowledging these biases, we can design instruments that reduce, though never eliminate, their impact. Patient-reported data will always be an approximation but by designing with cognitive reality in mind, we can make it a closer approximation.

Thank you for reading,

Mark Gibson, Leeds, United Kingdom

Nur Ferrante Morales, Ávila, Spain

September 2025