More work continues to enter the site even after principal investigator-dependent work is already waiting for the investigator’s time. As waiting work exceeds what available investigator bandwidth can clear, each additional item joins the same queue rather than increasing site throughput. Site workload and executable throughput have started to separate.
Principal investigator bandwidth matters because some site work cannot progress without that involvement. The exact form of the involvement may differ, but the capacity relationship remains the same. [1] Work reaches a point where principal investigator availability is required, and progression through that point is limited by the amount of investigator bandwidth available. Before that capacity is fully used, the site may absorb more dependent work without affecting throughput. At that limit, the same bandwidth becomes binding.
This distinction matters because rising site activity can look like evidence of remaining capacity but is not sufficient evidence when the additional work still depends on a principal investigator step that is already operating at its limit. The incoming volume can enter the system while completed progression remains bounded by the same investigator capacity.
The constraint sits inside the site
Once principal investigator bandwidth becomes binding, the visible effect appears at site level even though the limiting condition is narrower. Site throughput stops scaling with added activity because the principal investigator-dependent portion of the work cannot progress at a higher rate. That can make the site appear broadly constrained when the operative limit is at one required point in progression.
The distinction is important for interpretation. A site may show substantial activity across its workflow while the rate at which work progresses to completion changes little. That divergence does not, by itself, establish why throughput is constrained. The relevant evidence is whether the amount of work requiring principal investigator involvement has reached what available investigator bandwidth can process. Without that connection, a throughput ceiling can be observed without identifying the capacity that sets it.
This is also why a high workload alone is not enough to establish a bandwidth constraint. The binding condition appears only when the amount of work requiring principal investigator involvement reaches what available investigator capacity can process at the required rate. Before that point, bandwidth exists as a potential limit. After that point, it becomes an active limit on progression.
Activity can overstate executable capacity
The most consequential operational effect is a mismatch between apparent site capacity and executable throughput: site activity can continue to rise even when the principal investigator-dependent point cannot support additional progression. If activity is used as the primary signal of headroom, the site may appear able to absorb more work while the throughput-limiting step has no remaining capacity. [2]
Observed throughput presents a different interpretive limitation. Current output shows what the site is producing, but it does not reveal how close principal investigator bandwidth is to its limit. The same throughput level could be produced with unused investigator capacity or with investigator capacity fully consumed, so throughput alone cannot distinguish those conditions.
That means remaining capacity cannot be inferred from either activity or output in isolation. Leadership needs to understand the available principal investigator bandwidth relative to the amount of dependent work reaching it. The rate at which that work progresses through the investigator-dependent point then shows whether apparent headroom represents remaining capacity or whether that capacity is already exhausted.
The timing of the constraint also matters for interpretation. If visibility improves and waiting work becomes easier to see, the underlying capacity condition may be unchanged. Faster recognition of delayed progression does not mean the principal investigator ceiling has moved. Evidence of a changed condition requires a change in available bandwidth or in the amount of work reaching it.
What leadership can actually infer
For clinical operations leaders, the oversight question is whether expected site throughput fits within available principal investigator bandwidth. When the investigator-dependent step is binding, overall site activity cannot answer that question because activity may keep rising without showing that the site has matching executable capacity.
This changes how site performance should be read. Increasing activity does not necessarily mean more work can progress, and stable output does not necessarily indicate spare investigator capacity. Both observations are incomplete without visibility into the principal investigator-dependent work and the bandwidth available to process it.
The ceiling is also conditional rather than permanent; principal investigator bandwidth does not constrain every unit of site activity at all times. It defines the maximum rate of progression available through the work that depends on that bandwidth, and it becomes operationally decisive only when required load reaches that capacity. This preserves an important distinction between the existence of a finite resource and the point at which that resource limits throughput.
Beyond that point, the relevant capacity question becomes how much site throughput the available principal investigator bandwidth can support. Leadership therefore needs to assess whether that bandwidth can sustain the rate of dependent work required for expected site throughput.
Sources:
[1] International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. Guideline for Good Clinical Practice E6(R3). ICH Harmonised Guideline, final version adopted 6 January 2025; typographical corrections dated 24 October 2025. ICH E6(R3) final guideline
[2] Springer. Palvannan RK, Teow KL. Queueing for healthcare. Peer-reviewed article, Journal of Medical Systems, published online 8 May 2010; print 2012. PubMed-indexed article