From a Clinical Target to an Executable Rule
A target-based titration rule is a prespecified procedure for changing a patient's dose in response to measured outcomes. Its purpose is not merely to state a desired response, such as “reach adequate symptom control.” It must also say which patients the rule applies to, what measurements qualify for a decision, which direction the dose changes in each case, and what happens when the target or a safety boundary is reached.
The rule is one part of a regimen decision. The clinical team must choose a meaningful target and safety criterion, and the allowed dose options must be feasible for the formulation and protocol. The rule then makes the decision process reproducible: patients with the same defined inputs receive the same algorithmic action. A rule that leaves an input combination unanswered is not executable.
The example throughout this tutorial is hypothetical and illustrates specification, not a treatment recommendation. Suppose the protocol permits doses \(D=\{10,20,30\}\) mg once daily. Let \(R_k\) be the response measured at decision \(k\), in response units, where larger values represent greater benefit. The target is \(R^*=50\) response units. Let \(S_k\) be a safety score, in points, where larger values indicate greater safety concern; the prespecified safety limit is \(S_{\max}=8\) points.
Before applying any rule, the protocol must establish that the target is clinically interpretable, the measurement method and units are defined, and the safety measure and threshold are appropriate for the intended population. It must also specify the dose options and their order. This example assumes that 10, 20, and 30 mg are the only permitted doses, and that the response and safety values used for a decision are valid measurements collected under the protocol's assessment procedures.
Eligibility, Measurement, and Decision Priority
Eligibility conditions define who may enter the algorithm; they are not a substitute for the dose-action rules. For illustration, suppose entry requires adults aged 18 through 75 years, a specified stable background regimen, and no protocol-defined contraindication. A patient who does not meet an entry condition does not start this algorithm. The actual conditions must be written in operational terms so that two users can reach the same eligibility decision.
At a dose decision, both \(R_k\) and \(S_k\) must be available and valid. A value is valid only if it meets the protocol's measurement and quality requirements and is expressed in the specified units. If either value is missing or invalid, this example's rule keeps the current dose and makes no algorithmic adjustment pending a valid assessment. The assessment window and procedures for obtaining a valid measurement must be specified by the protocol; they are not inferred from the target itself.
Decision priority matters when more than one condition applies. In the illustrative rule, a safety score above the limit has priority over response: if \(S_k>8\), treatment and titration stop, regardless of \(R_k\). A score equal to 8 does not meet the “above the limit” stopping criterion. The protocol must separately provide appropriate clinical management; this tutorial specifies only the algorithmic consequence.
A Complete Rule With Dose-Boundary Actions
For valid measurements in an eligible patient, compare the response with the target and the current dose with the permitted dose limits. The dose levels are ordered, so the next higher or lower permitted dose is defined whenever one exists. The rule below also specifies what to do when it does not exist.
The response-based rows apply only after the patient has met the entry conditions, the current dose is one of the permitted levels, measurements are valid, and \(S_k\leq 8\). An ineligible patient does not enter the algorithm. A current dose outside the permitted set is a protocol discrepancy, not an invitation for the rule to guess a dose; it must be reconciled before applying the algorithm.
The response comparisons in this example use the measured values as recorded by the prespecified method, without an additional, unstated rounding step. A real protocol should state the precision and any rounding convention before comparisons are made. The safety comparison is likewise explicit: \(S_k=8\) is not a trigger, whereas any valid value greater than 8 is.
Worked Example: Increasing Below the Target
A hypothetical eligible patient is taking 20 mg. The valid assessment gives \(R_k=43\) response units and \(S_k=5\) points. Determine the action.
First check the safety condition: \(5\leq 8\), so the stopping criterion is not met. Both measurements are valid, and 20 mg is an allowed dose. Next compare response with target: \(43<50\), a shortfall of \(50-43=7\) response units. Since 20 mg is below the maximum of 30 mg, the next higher permitted dose is 30 mg.
The rule therefore increases the dose from 20 mg to 30 mg. It does not calculate a proportional increase from the 7-unit shortfall; the action is determined by the prespecified ordered dose options.
Worked Example: High Response at the Lowest Permitted Dose
A hypothetical patient is at 10 mg, the lowest permitted dose. The valid measurements are \(R_k=56\) response units and \(S_k=6\) points. What does the rule do?
The safety score satisfies \(6\leq 8\), so the safety stopping condition does not apply. The response is above target by \(56-50=6\) response units. Ordinarily, a response above target calls for a decrease, but there is no lower permitted dose below 10 mg. The rule explicitly assigns this boundary case: maintain 10 mg and stop titration at the lower dose limit.
Thus the action is executable even though the target is exceeded. It does not request an unavailable dose or silently substitute a different action. The patient remains at 10 mg under this example's specified rule; clinical interpretation of the result belongs to the protocol's broader decision process.
Worked Example: Safety Override and Missing Measurement
Consider two hypothetical decisions. In the first, a patient is taking 20 mg, has \(R_k=48\) response units, and has a valid \(S_k=9\)-point safety score. In the second, a patient is taking 20 mg, has a valid \(R_k=48\), but the safety score is unavailable.
First decision. The safety value is above the limit: \(9>8\). That condition has priority over the response shortfall, so treatment and titration stop. The rule does not increase the dose despite \(48<50\).
Second decision. The safety value is missing, so the measurements are incomplete. The rule maintains 20 mg and makes no algorithmic adjustment pending a valid safety assessment. It does not treat an unknown safety value as though it were acceptable.
Why the Rule Is Complete
A useful check is to verify that the decision cases neither conflict nor leave gaps. The safety trigger is checked first. If it is absent, invalid or incomplete measurements have a defined action. With valid measurements and acceptable safety, the response is either below, equal to, or above 50. For below-target responses, the current dose is either below 30 mg or is 30 mg. For above-target responses, it is either above 10 mg or is 10 mg. Each possibility has one specified action.
These response cases are mutually exclusive and cover every numerical response value. The dose-boundary checks then ensure that a requested increase or decrease is available. In particular, an above-target response at 10 mg is not sent to an undefined “lower dose” action: it maps to maintaining 10 mg and stopping titration. The same check is needed at the upper boundary for a below-target response.
If the patient fails an entry condition, do not initiate this titration algorithm.
If a valid safety score exceeds 8 points, stop treatment and titration as specified.
If response or safety information is missing or invalid, maintain the current dose and make no algorithmic adjustment pending valid information.
Use the target comparison to choose a direction, then check whether a permitted dose exists in that direction. At the target or a dose boundary, apply the stated maintenance and stopping action.
Diagnostics, Alternatives, and Limitations
A rule can be mathematically complete yet perform poorly in practice. Before use, check the written specification against a table of input combinations, including the safety threshold itself, values just above and below the response target, missing measurements, and both dose boundaries. Confirm that every row has one unambiguous action. Also check that staff can determine eligibility and measurement validity from available records, and that the response and safety units are consistent throughout.
Common warning signs include frequent decisions deferred for invalid measurements, many patients reaching a dose boundary without the target being met, and actions that do not correspond to the intended clinical priorities. Such patterns may indicate problems with measurement procedures, the selected target, or the dose options. They do not, by themselves, prove which component should change. Any revision should be justified and the revised rule documented before it is applied.
A target-based rule differs from a fixed-dose approach, which assigns a dose without using an individual response to choose the next dose. It also differs from discretionary titration, in which a clinician uses judgment without a fully specified mapping from inputs to actions. A fixed dose can be simpler and more predictable but may not respond to individual measurements. Discretion can accommodate context not represented in a simple rule but may reduce consistency across users. A target-based rule makes its stated inputs and actions reproducible, but only for the circumstances it actually specifies.
The rule also has limitations. A response target does not guarantee that increasing the dose will improve response, and a safety threshold does not capture every possible safety concern. The example uses a single response target and a single safety trigger; it does not establish that these values are clinically appropriate. Measurement error, changes over time, and response delays can complicate decisions. The assessment procedure and the clinical rationale for the target must therefore be considered alongside the algorithm, rather than assumed to be solved by writing a decision table.
Common Mistake
The same principle applies at the upper boundary: “increase if response is below target” is not executable at the maximum dose unless the protocol states what happens there. Boundary conditions are part of the rule, not exceptional details to be improvised later.
Check Your Understanding
For the hypothetical rule in this tutorial, answer each question, then compare your reasoning with the solutions below.
- A patient at 20 mg has valid values \(R_k=50\) response units and \(S_k=8\) points. What action applies?
- A patient at 30 mg has valid values \(R_k=44\) response units and \(S_k=4\) points. What action applies?
- A patient at 20 mg has valid values \(R_k=54\) response units and \(S_k=8.2\) points. Which condition determines the action?
- Why must the rule specify an action for an above-target response at 10 mg?
- A patient meets entry criteria but the response measurement is invalid; the safety score is valid and equals 7 points. What does this example's rule do?
Solutions. (1) The safety value is not above the limit because \(8\not>8\). The response equals the target, so maintain 20 mg and stop titration at target. (2) Safety is acceptable and \(44<50\), but 30 mg is already the maximum; maintain 30 mg and stop titration at the upper dose limit. (3) The safety override applies because \(8.2>8\); stop treatment and titration, regardless of the above-target response. (4) At 10 mg no permitted lower dose exists, so an unqualified instruction to decrease is not executable. This rule maintains 10 mg and stops titration at the lower dose limit. (5) The response and safety information are incomplete as a decision pair because the response is invalid. Maintain 20 mg and make no algorithmic adjustment pending a valid assessment.