---

Abstract #

The Heuristic Prediction Framework for Pharma (HPF-P) introduced two complementary constructs for evaluating decision quality in AI-augmented pharmaceutical portfolio management: the Decision Readiness Index (DRI), which quantifies information sufficiency, and the Decision Readiness Level (DRL), which assesses organizational maturity. While each metric addresses a distinct dimension of readiness, their isolated application leaves critical gaps — a high DRI score means little without the organizational capacity to act on available information, and a mature DRL offers no assurance that the underlying data is adequate. This article presents a unified assessment protocol that integrates DRI and DRL into a single composite framework, the Decision Readiness Assessment Protocol (DRAP), providing pharmaceutical portfolio managers with a holistic instrument for evaluating when and how to commit resources to AI-driven portfolio decisions.

graph LR
    A[Information
Sufficiency] -->DRI| C[DRAP
Composite Score]
    B[Organizational
Maturity] -->DRL| C
    C --> D{Decision
Gate}
    D -->Ready| E[Execute Portfolio
Decision]
    D -->Not Ready| F[Identify Gaps &
Remediate]

Introduction #

Pharmaceutical portfolio management operates under compounded uncertainty: therapeutic pipelines span decades, regulatory environments shift unpredictably, and market dynamics in regions like Ukraine introduce additional volatility through conflict, currency instability, and supply chain disruption. The HPF-P framework was developed to provide decision-makers with structured tools for navigating this uncertainty, drawing on principles from information theory and economic cybernetics^[2] to formalize what “readiness to decide” actually means.

The Decision Readiness Index (DRI)^[3] addresses the data dimension: given a specific portfolio decision, how complete, timely, and reliable is the available information? DRI operationalizes Shannon entropy as a measure of remaining uncertainty, producing a normalized score that indicates whether the information environment supports confident action. The Decision Readiness Level (DRL)^[4], by contrast, addresses the organizational dimension: does the institution possess the processes, expertise, and AI infrastructure to translate information into effective action? DRL adapts maturity model methodology — well-established in both defense technology assessment^[5] and enterprise AI deployment^[6] — to the specific demands of pharmaceutical decision-making.

The need for integration is straightforward. Consider a Ukrainian pharmaceutical distributor with access to comprehensive market data (DRI = 0.85) but lacking the analytical infrastructure to process it (DRL = Level 2). The high DRI creates a false sense of readiness; the low DRL means the organization cannot act on its informational advantage. Conversely, a multinational with mature AI pipelines (DRL = Level 4) operating in a data-scarce conflict zone (DRI = 0.35) faces the opposite problem — organizational capacity without adequate input. Neither metric alone captures the full picture.

This article formalizes the integration through DRAP, a composite protocol that maps DRI×DRL combinations to actionable decision gates. We present the mathematical foundations, define the assessment workflow, and illustrate application through a simulated case study grounded in Ukrainian pharmaceutical market conditions.

Theoretical Foundations of Integration #

Information-Theoretic Basis #

The DRI builds on Shannon’s entropy formulation, where decision uncertainty H(D) is computed across the probability distribution of portfolio outcomes. Recent work has reinforced the applicability of entropy-based methods to clinical decision-making under uncertainty^[7], demonstrating that information-theoretic measures can meaningfully capture the reduction in decision ambiguity as evidence accumulates. The American Academy of Actuaries (2025)^[8] has further validated entropy as a practical tool for measuring statistical bias in decision-relevant data, noting that “the higher the entropy, the more uncertainty in the system.”

DRI normalizes this measure against a theoretical maximum, producing a 0-1 scale where values above 0.7 indicate sufficient information for portfolio-level decisions. The normalization accounts for domain-specific factors: therapeutic area complexity, regulatory regime stability, and market data availability.

Maturity Model Basis #

DRL draws on established maturity assessment methodology, adapted from frameworks like the MITRE AI Maturity Model^[5], which defines 20 dimensions across five readiness levels for AI deployment capability. The pharmaceutical adaptation restructures these dimensions around portfolio-specific requirements: data pipeline maturity, model governance, clinical evidence integration, and regulatory compliance automation. Recent empirical validation of AI maturity models in manufacturing^[9] has confirmed that such frameworks reliably predict deployment success when dimensions are appropriately calibrated to domain context.

The five DRL levels — from Ad Hoc (Level 1) through Optimizing (Level 5) — correspond to qualitatively distinct organizational capabilities. The critical insight from 2026 maturity assessment literature^[10] is that “AI maturity — not experimentation — will separate winners from laggards,” a principle that applies directly to pharmaceutical portfolio management where the cost of premature or poorly-supported decisions is measured in billions.

The Integration Problem #

quadrantChart
    title DRI × DRL Decision Space
    x-axis "Low DRI" --> "High DRI"
    y-axis "Low DRL" --> "High DRL"
    quadrant-1 "Capable but Blind"
    quadrant-2 "Ready to Act"
    quadrant-3 "Double Deficit"
    quadrant-4 "Data Rich, Action Poor"

The fundamental challenge is that DRI and DRL operate on different scales (continuous vs. ordinal) and measure different constructs (information adequacy vs. organizational capacity). Simple aggregation — averaging, for instance — obscures critical failure modes. A pharmaceutical company with DRI = 0.9 and DRL = Level 1 averages to a superficially acceptable composite, masking the operational inability to execute on available data. The integration protocol must preserve the discriminative power of each metric while enabling unified decision gating.

DRAP: The Decision Readiness Assessment Protocol #

Architecture #

DRAP operates as a two-dimensional gating function that maps the (DRI, DRL) pair to one of four decision states:

1. Proceed (DRI ≥ 0.7, DRL ≥ 3): Both information and organizational capacity meet thresholds. The portfolio decision may advance with standard governance.

1. Conditional Proceed (DRI ≥ 0.7, DRL = 2) or (DRI 0.5-0.7, DRL ≥ 4): One dimension compensates for marginal adequacy in the other. Requires additional review and risk acknowledgment.

1. Remediate (DRI < 0.5, DRL ≥ 3) or (DRI ≥ 0.5, DRL ≤ 2): One dimension is clearly insufficient. The protocol identifies specific gaps and prescribes remediation steps.

1. Defer (DRI < 0.5, DRL ≤ 2): Both dimensions are inadequate. Decision postponement with structured improvement plan.

Formal Definition #

Let DRI ∈ [0, 1] represent the normalized information sufficiency score and DRL ∈ {1, 2, 3, 4, 5} represent the organizational maturity level. The DRAP composite function Φ is defined as:

Φ(DRI, DRL) = min(DRI / τI, DRLnorm / τ_O)

where τI = 0.7 is the information threshold, τO = 0.6 is the organizational threshold (corresponding to DRL Level 3 on a normalized 0-1 scale), and DRL_norm = (DRL – 1) / 4 maps the ordinal scale to [0, 1].

The minimum operator ensures that the composite score is bounded by the weakest dimension — reflecting the practical reality that decision quality is limited by the most deficient input. This aligns with the multi-criteria decision analysis principle^[11] that health technology assessment must account for all value dimensions simultaneously, not just aggregate scores.

flowchart TD
    A[Collect DRI Score] --> B[Assess DRL Level]
    B --> C{Φ ≥ 1.0?}
    C -->Yes| D[PROCEED
Standard Governance]
    C -->No| E{Φ ≥ 0.7?}
    E -->Yes| F[CONDITIONAL
Enhanced Review]
    E -->No| G{min dimension?}
    G -->DRI| H[REMEDIATE: Data
Gap Analysis]
    G -->DRL| I[REMEDIATE: Capability
Development Plan]
    G -->Both| J[DEFER
Structured Improvement]

Assessment Workflow #

The DRAP assessment follows a structured protocol:

Phase 1 — DRI Computation. For each pending portfolio decision, the DRI module ingests available data sources: market analytics, clinical trial databases, regulatory filings, pharmacoeconomic models, and competitive intelligence. Shannon entropy is computed across the decision variable space, normalized against the domain-specific theoretical maximum. The module outputs both the aggregate DRI score and a decomposition by data category, enabling targeted remediation when scores are insufficient.

Phase 2 — DRL Evaluation. The DRL assessment surveys five capability dimensions: (a) data infrastructure and pipeline maturity, (b) analytical model governance, (c) human-AI decision integration, (d) regulatory compliance automation, and (e) continuous learning and model evolution. Each dimension is scored on the 1-5 maturity scale using structured rubrics adapted from the MITRE framework. The aggregate DRL is the minimum across dimensions — again applying the weakest-link principle.

Phase 3 — DRAP Gating. The composite function Φ is computed, and the decision is assigned to one of four states. For Conditional Proceed and Remediate outcomes, the protocol generates specific recommendations based on the dimensional decomposition from Phases 1 and 2.

Phase 4 — Longitudinal Tracking. DRAP scores are logged over time, enabling trend analysis. Organizations can track whether remediation efforts are improving readiness, identify recurring capability gaps, and benchmark against industry peers.

Case Study: Ukrainian Pharmaceutical Distributor #

To illustrate DRAP application, we construct a simulated scenario based on conditions documented in the environmental entropy analysis of Ukrainian pharmaceutical markets^[12].

Scenario #

A mid-sized Ukrainian pharmaceutical distributor is evaluating a portfolio rebalancing decision: shifting procurement allocation from generic cardiovascular drugs toward oncology biosimilars, driven by changing demographic demand patterns. The decision involves committing approximately $12M in procurement contracts over 18 months.

DRI Assessment #

The DRI module evaluates five data categories:

Data Category	Entropy (bits)	Max Entropy	Normalized Score
Market demand data	2.1	3.5	0.60
Regulatory status	1.0	2.8	0.64
Supply chain reliability	3.2	3.8	0.16
Clinical evidence base	0.8	3.0	0.73
Competitive landscape	1.9	3.2	0.41

The supply chain dimension reflects wartime disruption — logistics data is unreliable, with frequent route changes and inventory uncertainty. The aggregate DRI = 1 – H̄/H_max = 0.51, placing the decision below the 0.7 threshold. The decomposition clearly identifies supply chain intelligence as the primary deficit.

DRL Assessment #

DRL Dimension	Level	Rationale
Data infrastructure	3	Modern ERP with API integrations, but limited real-time analytics
Model governance	2	Ad hoc model validation, no formal MLOps pipeline
Human-AI integration	3	Decision support dashboards in use, but limited trust calibration
Regulatory compliance	3	Automated filing for standard submissions, manual for complex cases
Continuous learning	2	No systematic model retraining or drift detection

Aggregate DRL = min(3, 2, 3, 3, 2) = Level 2. The weakest-link principle highlights model governance and continuous learning as the binding constraints.

DRAP Result #

Φ(0.51, 2) = min(0.51/0.7, 0.25/0.6) = min(0.73, 0.42) = 0.42

Decision State: REMEDIATE (DRL-bound)

The protocol recommends:

1. Establish formal MLOps pipeline for model governance (target DRL dimension 2 → Level 3)
2. Implement model drift detection for demand forecasting (target DRL dimension 5 → Level 3)
3. In parallel, improve supply chain intelligence through alternative data sources (satellite logistics tracking, cross-border shipment databases) to address the DRI supply chain gap

After remediation, projected Φ(0.62, 3) = min(0.89, 1.0) = 0.89 → Conditional Proceed, achievable within an estimated 4-6 months.

Connections to Self-Improving Systems #

The DRAP protocol embodies a principle central to research on formal self-improvement in AI systems. Schmidhuber’s Gödel Machine concept^[13] — a self-referential system that modifies its own decision policy only when it can prove the modification improves expected utility — provides a theoretical anchor for DRAP’s gating mechanism. The requirement that both DRI and DRL must meet thresholds before a decision proceeds is analogous to the Gödel Machine’s proof obligation: the system does not act until it can demonstrate sufficient grounds for action.

This connection is not merely metaphorical. In the HPF-P platform architecture^[14], the DRAP gating function is implemented as an automated checkpoint in the decision pipeline. When new data arrives (improving DRI) or organizational capabilities are upgraded (improving DRL), the system automatically re-evaluates pending decisions against their DRAP thresholds — a practical instantiation of the self-improving principle where the system’s decision boundaries evolve with its informational and operational state.

Implications for Regulatory Frameworks #

The integration of DRI and DRL has implications beyond organizational decision-making. The FDA’s 2026 guidelines on AI in drug development^[15] emphasize the need for transparent, auditable decision processes when AI systems influence regulatory submissions. DRAP provides exactly this transparency: every portfolio decision carries a documented assessment of both information adequacy and organizational capability, creating an audit trail that regulators can inspect.

Niazi (2026) notes in a critical review of the FDA’s draft guidance that “regulatory agencies worldwide are developing frameworks to guide the responsible implementation of AI-driven drug development.” DRAP aligns with this trajectory by providing a standardized, quantitative protocol that organizations can present during regulatory review to demonstrate decision rigor.

The emerging literature on AI-enabled regulatory ecosystems^[16] further supports the need for integrated readiness assessment, arguing that fragmented evaluation of AI capabilities — whether focused solely on model performance or solely on organizational governance — produces incomplete regulatory pictures. DRAP’s dual-axis assessment addresses this gap directly.

Limitations and Future Work #

Several limitations warrant acknowledgment. First, the DRL assessment relies on structured rubrics that, while empirically grounded, involve subjective judgment in scoring. Inter-rater reliability studies across pharmaceutical organizations of varying size and sophistication are needed. Second, the minimum operator in both DRL aggregation and DRAP composite computation is conservative by design — it may underweight compensatory effects where strength in one dimension partially offsets weakness in another. Alternative aggregation functions (geometric mean, weighted minimum) deserve empirical comparison.

Third, the case study presented here is simulated. Prospective validation with real pharmaceutical decision-makers — comparing DRAP-guided decisions against traditional portfolio management outcomes — remains the critical next step for establishing protocol validity. The experimental validation work on HPF multi-strategy optimization^[17] provides a methodological template for such studies.

Finally, the current DRAP formulation treats DRI and DRL as independent dimensions. In practice, organizational maturity affects data collection capability (a high-DRL organization generates better data, improving DRI), and information availability influences process development (organizations invest in capabilities when they see data-driven opportunities). Modeling these feedback loops — potentially through dynamic systems methods — would produce a more realistic assessment protocol.

Conclusion #

The Decision Readiness Assessment Protocol (DRAP) unifies the informational and organizational dimensions of pharmaceutical portfolio decision-making under a single, auditable framework. By combining the entropy-based DRI with the maturity-grounded DRL through a minimum-operator gating function, DRAP ensures that decisions proceed only when both prerequisite conditions are met. The protocol’s four-state output — Proceed, Conditional Proceed, Remediate, Defer — provides actionable guidance that goes beyond a simple pass/fail, directing remediation efforts toward the specific dimensions that limit overall readiness.

For the HPF-P framework, DRAP represents a critical integration layer. The individual metrics — DRI for information, DRL for capability — are necessary but insufficient components of a complete decision readiness assessment. Their unification through DRAP completes the framework’s core assessment module, enabling the next phase of development: empirical validation with practicing pharmaceutical portfolio managers operating under real-world uncertainty conditions.