Automated Insulin Delivery (AID)

What AID is and why it matters

Automated insulin delivery (AID) is the current regulatory and commercial term for systems that combine three core elements: a continuous glucose monitor (CGM), an insulin pump, and a control algorithm that adjusts insulin delivery—usually basal insulin—based on glucose data. In the clinical literature, many of the approved systems are called hybrid closed-loop systems because they automate basal delivery but still require the person to announce meals and deliver meal-related boluses.

A foundational granted patent in this space is US 6,558,351 B1, titled “Closed loop system for controlling insulin infusion”, granted 2003-05-06, assigned to Medtronic MiniMed, Inc. [S1]. The patent is evidence of recorded claims to a closed-loop architecture, not evidence that any specific commercial device achieved a particular glucose outcome.

The relevant FDA regulatory record for one historically important system is the Medtronic MiniMed 670G System PMA, P160017, with an FDA approval date of 2016-09-28 [S4]. The MiniMed 670G represented a regulatory milestone: the first FDA-approved hybrid closed-loop insulin delivery system. That approval is a regulatory fact, not a comparative performance claim.

This note treats the category name “automated insulin delivery” carefully. In neither the 2016 FDA approval nor the randomized evidence cited below is the system fully autonomous. Patients still make meal-bolus decisions and must respond to alarms. The world “automated” is a category label; the verified evidence is for hybrid closed-loop operation.

Source map

The hard tier rules are respected as follows:

The strongest evidence for technical/IP claims comes from the granted patent and the FDA PMA. The strongest evidence for clinical performance claims comes from the peer-reviewed randomized trials. A granted patent is a record of claimed invention; a peer-reviewed trial is a record of measured outcomes. They answer different questions.

Anatomy of an AID system

The load-bearing technical claim—that an AID system integrates a CGM, pump, and controller to adjust insulin infusion—is cross-verified by multiple independent sources:

• The granted patent US 6,558,351 B1 describes a closed-loop infusion system with a sensor for determining a body characteristic, an infusion pump, and a controller that adjusts infusion rate [S1].
• The FDA PMA record for the MiniMed 670G System identifies the device as an insulin pump system with CGM and hybrid closed-loop functionality [S4].
• The 2019 New England Journal of Medicine randomized trial of a closed-loop system used a t:slim X2 insulin pump with Control-IQ technology and a Dexcom G6 CGM [S2].

The systems therefore share a common architecture:

1. CGM — measures glucose in interstitial fluid.
2. Insulin pump — delivers rapid-acting insulin via subcutaneous cannula.
3. Algorithm — computes basal-rate changes, suspension, or increased delivery based on sensor values.
4. User interface — enables meal boluses, alarm acknowledgment, and mode changes.

This architecture is the central value proposition of AID: it moves basal-insulin decisions from a static preprogrammed basal rate to a continuously adjusted rate. But the system is not a closed-loop “artificial pancreas” that requires no user input; the clinical trial protocol in the strongest efficacy paper explicitly required meal boluses [S2].

Regulatory milestone and clinical evidence

4.1 FDA approval of MiniMed 670G

The FDA PMA record P160017 identifies the MiniMed 670G System as approved, with an approval date of 2016-09-28 [S4]. This is a Tier 1 regulator record. It establishes that the FDA found the device approvable under the PMA pathway. It does not by itself provide comparative effectiveness numbers.

Early safety evidence for the MiniMed 670G was published in JAMA in 2016 as “Safety of a Hybrid Closed-Loop Insulin Delivery System in Patients With Type 1 Diabetes” [S3]. That report described the system as hybrid closed-loop, reinforcing that the approved system was not fully autonomous.

4.2 Randomized trial: Control-IQ

The highest-grade efficacy evidence cited here is Brown et al., Six-Month Randomized, Multicenter Trial of Closed-Loop Control in Type 1 Diabetes, N Engl J Med 2019;381:1707–1717 [S2].

Key verified results from that paper, as relevant to this note:

• Time in range 70–180 mg/dL during 6 months:
• closed-loop group: 71% ± 12%
• control group: 59% ± 14%
• adjusted mean difference: 11 percentage points (95% CI, 9 to 13; P<0.001)
• Mean glucose at 6 months:
• closed-loop group: 154 ± 18 mg/dL
• control group: 173 ± 22 mg/dL
• adjusted mean difference: −18 mg/dL (95% CI, −27 to −10; P<0.001)
• HbA1c at 6 months:
• closed-loop group: 7.06% ± 0.79%
• control group: 7.39% ± 0.87%
• adjusted mean difference: −0.33 percentage points (95% CI, −0.55 to −0.11; P=0.003)

These results are from a single randomized trial and are therefore tagged [SINGLE-SOURCE] for the specific efficacy point estimates. They are the primary clinical feasibility evidence in this note. A second independent trial in children, Breton et al., A Randomized Trial of Closed-Loop Control in Children with Type 1 Diabetes, N Engl J Med 2020;383:836–845 [S10], supports the broader pediatric extension, but this note does not use it as a duplicate of the Brown point estimates.

4.3 Why the FDA approval and NEJM paper are different evidence types

• FDA PMA = regulatory clearance of safety and effectiveness for a specific device under conditions of approval. It does not independently produce the effect sizes; it is the governmental accepting authority.
• NEJM RCT = measured patient-level outcomes in a defined protocol, producing effect sizes with confidence intervals.
• Granted patent = intellectual-property claims. A patent does not show a device is safe, effective, or commercially viable.

Worked calculation: incremental time-in-range

This calculation converts the Brown et al. adjusted TIR difference into hours per day and annualized hours. It is a derived estimate, not a patient-level guarantee.

Formula:

\[

TIR_{\text{additional hours/day}} = 24\ h \times \Delta TIR

\]

Where \(\Delta TIR\) is the adjusted difference in proportion of time in range between closed-loop and control arms.

Inputs from S2:

• \(\Delta TIR = 0.11\) (11 percentage points; 95% CI, 0.09 to 0.13)

Central calculation:

\[

24\ h \times 0.11 = 2.64\ h/day

\]

Sensitivity bounds using the 95% CI:

• Lower bound: \(24 \times 0.09 = 2.16\ h/day\)
• Upper bound: \(24 \times 0.13 = 3.12\ h/day\)

Annualized central estimate:

\[

2.64\ h/day \times 365\ days = 963.6\ h/year

\]

Annualized sensitivity bounds:

• Lower: \(2.16 \times 365 = 788.4\ h/year\)
• Upper: \(3.12 \times 365 = 1,138.8\ h/year\)

Interpretation:

In the Brown trial, the closed-loop group spent approximately 2.6 additional hours per day in the 70–180 mg/dL range compared with the control group, with a plausible range of 2.2 to 3.1 hours per day. This is a population-level mean difference, not a guarantee for any individual user.

This calculation is labeled [SINGLE-SOURCE] for the underlying TIR estimate because the Brown et al. point estimate is not cross-verified by a second independent study in this note.

Parameter table

Patents versus performance

The single granted patent cited here, US 6,558,351 B1, is not a vintage issue — it is a foundational early closed-loop control patent from the MiniMed/Medtronic lineage [S1]. It is valid as evidence that a closed-loop architecture was described and claimed. It is not evidence that:

• The MiniMed 670G or any modern AID device achieved specific time-in-range results.
• The algorithm in the patent is the same one used in current commercial systems.
• The patented system was economically viable.

This distinction is critical for moat/feasibility analysis. A patent can support an IP-moat argument only if the claims map to the commercial product. In this note, I do not claim that the 2003 patent claims are infringed or practiced by any specific current device; that mapping would require a claims chart against a current product’s firmware and system architecture, which is not available from the cited public sources.

Contradictions, source discrepancies, and single-source flags

Contradiction 1: “Automated” vs “hybrid closed-loop.”

The category term “automated insulin delivery” appears in commercial language, but the strongest clinical and regulatory sources use the phrase “hybrid closed-loop.” The Brown trial required meal boluses by protocol [S2]; the 2016 JAMA report labels the MiniMed system as a “hybrid closed-loop insulin delivery system” [S3]; and the FDA PMA page identifies a PMA-approved MiniMed 670G System [S4]. This is a labeling nuance, not a fraud claim. It matters because a fully automated system that requires no meal input would be a different product than the one with verified trial evidence.

Contradiction 2: Patent breadth vs clinical evidence.

The patent [S1] predates the pivotal trials by more than a decade and is broad in architecture. The clinical papers measure outcomes of later, separately regulated devices. A patent is a legal document, not a performance document. Any investment argument that treats the patent alone as proof of clinical viability is unsupported.

Single-source flags:

• The TIR and HbA1c point estimates from Brown et al. are [SINGLE-SOURCE] in this note.
• The early MiniMed safety report [S3] is another lineage but does not independently replicate the Brown effect estimates.
• The FDA PMA approval date [S4] is not a comparison metric; it is a regulatory milestone.

Company-filing cross-check limitation:

I have not retrieved and quoted a specific 10-K risk-factor sentence in this pass. The SEC EDGAR indexes for Medtronic, Tandem, Insulet, DexCom, and Abbott are cited as primary retrieval points [S5–S9]. Any claim about exact risk-factor language must be verified against the actual SEC filing text; if quoted without that retrieval, it is needs primary source validation.

Unit economics / business model

AID monetization occurs at multiple layers:

1. Durable pump hardware — one-time sale or amortized rental through insurance.
2. Disposable pump consumables — infusion sets, cartridges, or tubeless pods that are replaced every 2–3 days.
3. CGM sensors and transmitters — disposable sensors replaced every 7–14 days; transmitters replaced periodically.
4. Control algorithm — currently embedded in the pump/interoperable system rather than sold as a separate software line-item subscription in the verified sources.
5. Service and support — training, cloud data platforms, and healthcare-provider dashboards.

Who captures the value? The pump manufacturers—Medtronic, Tandem, Insulet—capture the pump and disposable-pump revenue; the CGM makers—DexCom, Abbott—capture sensor revenue. The algorithm value accrues to the pump system, because the algorithm is not sold separately as a pure software product in the sources cited here.

Margins are not separately disclosed for the AID algorithm in the SEC indexes cited here. Pump and CGM companies report overall gross margins, but this note does not quote a specific margin because a tier-1 verified disaggregated AID margin was not retrievable in this pass. Any specific gross-margin number for a specific ticker is needs primary source validation.

Disconfirming Evidence

The AID thesis would be materially weakened or falsified if:

1. Independent randomized trials show no statistically significant improvement in time-in-range or HbA1c versus sensor-augmented pump therapy.
2. Severe adverse events—diabetic ketoacidosis, severe hypoglycemia, or malfunction-related insulin over-delivery—show a materially worse profile than control in a subsequent trial.
3. FDA issues a recall, safety alert, or new regulatory condition that suspends or restricts current AID indications in a major market.
4. Payors withdraw or narrow coverage for AID pumps or CGM sensors, collapsing the reimbursement pathway for the durable hardware and recurring disposables.
5. CGM accuracy degrades below the minimum required for safe closed-loop control, which would invalidate the core sensor-dependent algorithm assumption.
6. A granted patent claim is successfully challenged or narrowed during an inter partes review or litigation, undermining the IP-moat argument for a specific product family.