Complements or Substitutes? Technology Adoption and Clinical Care Utilization: Evidence from Automated Insulin Delivery
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Complements or Substitutes? Technology Adoption and Clinical Care Utilization".
Warren: Here's the gist of the paper. This study examines whether automated insulin delivery (AID) technology reduces or increases demand for professional care among adults with type 1 diabetes in four specialist clinics of the Italian National Health Service, where 283 users adopted AID.
Ruth: After adjustment, AID adoption is associated with increased routine outpatient engagement.
Warren: Specifically, the estimated effect on the probability of a diabetologist visit rises from 16.3 percentage points one semester after adoption to 39.3 percentage points four semesters later, compared to an observed visit probability of 55.2% in the reference semester before adoption.
Ruth: The findings are consistent with a task-based view of medical technology, suggesting that automation substitutes for routine dosing decisions but complements clinical labor by generating new interpretive, adjustment, and supervisory tasks.
Paper summary: Ruth: So we're looking at this paper titled "Complements or Substitutes? Technology Adoption and Clinical Care Utilization: Evidence from Automated Insulin Delivery." It’s about whether automated insulin delivery technology either reduces or increases the need for professional care among adults with type one diabetes <ref:2609.12976#pg1,adults with type 1 diabetes>.
Warren: Right, it looks like the study focuses on eighteen hundred people with type one diabetes in four specialist clinics in Italy, and two hundred eighty-three of those people adopted this AID technology <ref:2609.12976#pg1>. The main finding they're pointing to is that after accounting for pre-adoption trends, adopting AID is linked to increased routine outpatient engagement.
Iris: That sounds like a pretty significant finding because it moves beyond just looking at how much better the blood sugar control gets with the technology and starts looking at how it changes the structure of care itself. The paper claims this is consistent with a task-based view of medical technology, where automation takes over routine dosing but also creates new tasks for doctors to do, like interpreting data or supervising things.
Ruth: Exactly. The paper suggests that AID isn't just replacing one thing; it’s shifting the labor required in the clinic to more interpretive and supervisory roles. They are testing this idea across four semesters after adoption, looking at visits to a diabetologist.
Warren: And the numbers they present show this increase is conditional on how you look at the timeline. After one semester of adoption, they see an estimated rise of sixteen point three percentage points in the probability of a diabetologist visit compared to before it started.
Iris: But then it rises again four semesters later to thirty-nine point three percentage points, which is a substantial jump when you consider that their baseline visit probability before adoption was observed at fifty-five point two percent. That's a big shift in engagement.
Ruth: It seems the central argument here is that AID adoption doesn't necessarily reduce the specialist involvement; instead, it seems to prompt more specialist contact later on, even if the initial effect isn't immediately clear. They found that the probability of at least one diabetologist visit rises post-adoption because of this trend extrapolation assumption they have to make.
Warren: So you’re saying this paper is testing if the economic consequences depend on how technology reshapes demand for professional care, not just the direct cost of the device or immediate health outcomes. It's about how things change over time.
Iris: And from an econometric perspective, they acknowledge that this result is sensitive to that extrapolation assumption they make about what would have happened without AID adoption. They’re cautious there, saying the visit estimates should be interpreted conditionally on that restriction.
Ruth: That makes sense. It sounds like the main contribution of this paper is providing evidence on routine outpatient engagement as an effect of AID adoption, which isn't something that gets much attention in existing literature on this topic and complements earlier work they’ve done on glycaemic outcomes and reducing acute care use.
Warren: And another point they’re making is extending the task-based framework to patient-facing technologies where the clinician supplies the input for those new tasks, which is a broader way of looking at automation. It suggests that technology can shift resources toward sustained specialist involvement.
Iris: I think what's really interesting for us is how they frame this trade-off between outcome improvement and care demand change. They argue that these additional outpatient care needs might be worth it if they improve the blood sugar control or prevent acute events and long-run complications.
Ruth: But they're careful to stress that these additional outpatient care needs aren't automatically beneficial; more specialist contact isn't inherently good, and it could also mean more resources are being used. They conclude that because of this, savings in routine specialist care shouldn't be assumed when we look at the overall resource implications of the technology.
Warren: So what this means for us on air is that we shouldn't just assume saving money on routine specialist care just because a new technology is in use; we have to look at how utilization changes, not just whether a patient gets better.
Iris: And they also flag a limitation about the extrapolation assumption. They admit that because of this restriction, the results aren't completely certain and need careful consideration when interpreting the rise in visit probability they found.
Ruth: So, to wrap up on this paper, "Complements or Substitutes? Technology Adoption and Clinical Care Utilization: Evidence from Automated Insulin Delivery," it shows that automation isn't just about replacing a task; it’s about creating new clinical demands for doctors to handle.
Warren: And the main implication is that the economic assessment of this technology has to include how it changes the demand for professional care, not just device costs or immediate health metrics. It forces us to look at healthcare utilization over time.
Iris: It’s a reminder that even when a technology seems efficient in one area, its impact on the entire care system—the specialist involvement—needs careful measurement and context.
Ruth: We'll stop there for this discussion on the paper, but next up we're looking at something completely different.
Conclusion: Ruth: So, we've seen how automated insulin delivery adoption seems to change routine visits for type one diabetes patients in Italy by looking at some specialist clinic data.
Warren: Yeah, and the core of this paper is asking whether this technology just takes over routine dosing decisions or if it actually changes what doctors have to do with those patients.
Iris: The authors are really focused on that task-based view, suggesting that instead of replacing a task, the technology creates new ones for clinical labor.
Ruth: Right, they're looking at how the adoption of AI shifts the workload—from just dosing to interpreting data and supervising. It’s not just about what the machine does now; it's about what happens in those next steps.
Warren: And a key point they make is that this isn't necessarily about fewer specialist visits overall, but rather how those visits change over time after the adoption takes hold.
Iris: They found that while there is an increase in routine outpatient engagement, it’s conditional on a few assumptions about how trends extrapolate forward in time. That makes the numbers they show pretty tricky to take as a definitive statement right away.
Ruth: Exactly, so what this means for us is that we can't just look at immediate health improvements and assume the cost structure stays the same; we have to think about whether more specialist contact is actually helping or just adding more resource demands.
Warren: It forces us to assess the whole system utilization, not just whether a patient’s blood sugar numbers get better.
Iris: And their main contribution is showing that this effect on routine engagement hasn't been studied as much before, which complements earlier work on acute care use and glycaemic control.
Ruth: So the authors are suggesting we need to look at healthcare utilization changes over the long run when evaluating these kinds of medical technologies.
Warren: That leads us nicely into what happens when we look at other areas, like how these technologies interact with different patient groups or different types of care settings in the future.
University of Bologna
econ.GN, q-fin.EC
Submitted: 2026-09-11
Updated: 2026-11-14
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 82/100
The gist: This study examines whether automated insulin delivery (AID) technology reduces or increases demand for professional care among adults with type 1 diabetes in four specialist clinics of the Italian…
Key concepts
- Automated Insulin Delivery (AID)
- AID technology automates insulin delivery for type 1 diabetes patients. The study looks at how using this automated system changes the need for regular visits to a specialist like a diabetologist.
- Task-Based View of Medical Technology
- This view suggests that new medical technology doesn't just replace old tasks but changes the work involved. AID substitutes routine dosing decisions but adds new tasks requiring clinical interpretation, adjustment, and supervision from healthcare providers.
- Routine Outpatient Engagement
- This refers to the frequency of regular visits patients make to their specialist clinic for ongoing care. The study found that AID adoption is linked to an increase in this type of routine outpatient engagement.
Terminology
Summary
This study examines whether automated insulin delivery (AID) technology reduces or increases demand for professional care among adults with type 1 diabetes in four specialist clinics of the Italian National Health Service, where 283 users adopted AID. p. 0 After adjustment, AID adoption is associated with increased routine outpatient engagement. p. 0 Specifically, the estimated effect on the probability of a diabetologist visit rises from 16. p. 03 percentage points one semester after adoption to 39. p. 03 percentage points four semesters later, compared to an observed visit probability of 55. p. 02% in the reference semester before adoption. p. 0
The findings are consistent with a task-based view of medical technology, suggesting that automation substitutes for routine dosing decisions but complements clinical labor by generating new interpretive, adjustment, and supervisory tasks. p. 1 The paper makes two primary contributions: first, it provides evidence on how AID adoption affects routine outpatient engagement—a margin not addressed by existing AID literature—which complements previous findings on improved glycaemic outcomes and reductions in acute/inpatient utilization. p. 5 Second, it extends the task-based framework of automation to a patient-facing therapeutic technology where clinical input is supplied by clinicians. p. 6
The most robust conclusion is that AID adoption does not reduce recorded specialist involvement. p. 5 While the probability of at least one diabetologist visit rises post-adoption, this increase is conditional on the trend-extrapolation assumption. p. 0 The estimated effect on the probability of a visit is small and statistically insignificant at adoption, rising to 16. p. 03 percentage points one semester later and reaching 39. p. 53 percentage points four semesters later (with a joint test rejecting the null hypothesis of no effect at p = 0. p. 5044). This pattern is consistent with a task-based reinstatement mechanism, but it does not distinguish reinstatement from other mechanisms capable of generating greater clinical engagement after adoption. p. 6
The resource consequences of AID extend beyond acquisition costs; additional specialist time, monitoring, and follow-up may be required. p. 6 However, the study concludes that these additional outpatient care needs may be worthwhile if they improve glycaemic outcomes or prevent acute events and long-run complications. p. 5 Crucially, the findings speak to healthcare utilization rather than welfare: more specialist contact is neither necessarily beneficial nor necessarily inefficient. p. 4 Therefore, savings in routine specialist care should not be assumed when assessing the overall resource implications of the technology. p. 0
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Develop an AI system capable of estimating whether medical technology adoption substitutes for or complements professional clinical care by analyzing routine outpatient engagement, as demonstrated by AID adoption data. This system can determine if automation
does not reduce recorded specialist involvement,
which is consistent with atask-based reinstatement mechanism.
-
Enhance the system's ability to predict the dynamic path of healthcare utilization following technology adoption by incorporating trend adjustment based on event time extrapolation. The improved system can generate estimates that are
conditional on the trend-extrapolation assumption,
providing a more accurate view of how engagement changes over time, such as rising from 16.3 percentage points one semester after adoption to 39.3 percentage points four semesters later for diabetologist visits. -
Implement a mechanism within the AI to distinguish between task reinstatement and scale response in the demand for professional care by analyzing the decomposition of clinical demand into task content and scale terms. This allows the system to assess whether
automation mostly reallocates existing management tasks to the device, with little new clinical work
or ifinterpretation, adjustment, training, and supervision tasks AID creates are large enough
to cause a rise in routine specialist contact.
Related papers
- Huge If True: Learning from the Size of the Surprise
- The Rebundling of Freelance Work in the AI Age
- Dynamic investment and coordination in teams: Theory and experimental evidence
- Does Frontier AI Strengthen Cyber Defense? Evidence from Project Glasswing
- Generative AI at Work
- What is Work For? Labor, Politics, and Meaning in the Age of AI