T title of the article by Mar and others contains a term that is probably new to many readers. They may be pleased to learn that the term Budget Impact Analysis (BIA) means exactly what it says. A recent report from a task force under ISPOR (International Society for Pharmacoeconomics and Outcomes Research) describes the underlying philosophy and offers a guideline for good BIA practice and reporting. A BIA supplements other types of health-economic analyses by describing the likely future ‘‘trajectory of spending’’ in the wake of introducing a new drug or changing a reimbursement scheme, and so on. The ideal BIA is transparent and comes with a built-in sensitivity analysis, most likely in the form of a spreadsheet that allows a health plan manager or governmental health service to modify the scenario and observe the resulting effect on the cash-balance consequences of the envisaged change. As stressed in Neumann’s commentary, a BIA serves the very humble, pragmatic goal of examining ‘‘affordability’’: Will the existing insurance premiums or taxation be able to cover the expected change in prescription practice? Although it does enter into the calculation that a more effective drug may lead to a reduced need for hospital admissions and other types of care, no attention is paid to the fact that the patient will have a better life. In sum, there is no place for health benefits or equity issues, no optimality criteria, no arena for tradeoffs. As a utility devotee, I prefer to view these matters thus: A cost-effectiveness analysis (CEA) or similar traditional health-economic analysis takes the casedefining event as given, be it an infarction or a broken leg, and tries to model expenditures and patient health diary forward in time until recovery or death. Imagine superimposing such individual case histories on the calendar, thereby creating what could be described as a stack of time-shifted cases. Suppose one reads the stack cross-sectionally. Then total monthly expenditures may be calculated, month by month, as well as the monthly total health-related utility gain (monthly total QALY production, if you prefer). Depending on perspective, the patient’s return to work and tax-paying may be included in the balance. In addition to money and well-being, one may also project the need for doctors, nurses, and equipment. Whether one’s predominant focus is longitudinal or cross-sectional, one may apply the model to alternative regimens in order to compare them in the light of agreed optimality criteria. When viewed within this framework, BIA is little more than the expenditures layer peeled off from the remainder of the stack. However, BIA also possesses some features of its own: It must pay attention to epidemiological and demographic trends; to case-to-case dependencies (such as contagion); to the rate at which a new drug will gain acceptance from prescribers; to market features, including competitors’ pricing decisions; and to the probable time it will take before a still more effective treatment becomes available. When a pharmaceutical company tries to ‘‘sell’’ a new drug to the relevant authority, it will glorify benefits and understate the rate of acceptance and the costs associated with side effects and other bycosts. It will also try to set the price of its product high enough to make a profit but low enough to obtain approval. Look at this from the authority’s side of the table, and you have a game-theoretic situation. Mar and others use Markov modeling. To me, the proper way to model a disease such as stroke is to use continuous time, with states that have a duration (hence a prevalence in the stack) and state transitions that are point-in-time events governed by transition rates (hence incidence). A stroke, or recurring stroke, is a point event that carries a monetary cost for acute treatment and moves the patient to a post-stroke state, or post-second-stroke state, characterized by a state-specific monthly cost of continued medication and by state-specific transition rates pertaining to subsequent events, which in turn trigger transition into a next state. States and state transitions both carry a cost. In my opinion, a discrete-time model as used by Mar and others is not just an innocent approximation. DOI: 10.1177/0272989X08321903