Why State Definitions Matter
A person may need assistance with daily activities without meeting the eligibility test in a long-term care contract. Conversely, a contract may treat severe cognitive impairment as qualifying even when the person can still perform several physical activities independently. A model that treats every diagnosis or care need as benefit eligibility can therefore misstate which state a person occupies.
The first task is to define states using observable conditions and explicit contract rules. For this tutorial, all product terms are hypothetical. The eligibility tests below illustrate how to translate contract language into state definitions; they are not descriptions of any particular policy.
Consider a hypothetical contract that regards a person as meeting its care trigger if either (1) a qualified professional certifies that the person cannot perform at least two of six specified activities of daily living, with the limitations expected to last at least 90 days, or (2) a qualified professional certifies severe cognitive impairment requiring substantial supervision. A diagnosis alone does not satisfy either test. The precise tests in an actual contract may differ.
In this example, “care-needing” means that the stated contractual trigger is met. It does not mean that a claim has necessarily been paid. A waiting period, documentation process, benefit limit or other payment condition may affect whether and when money is payable. Those payment conditions should not be silently built into the definition of a clinical state.
Four Useful State Definitions
A basic model may distinguish healthy, care-needing, recovery and death. These names are convenient, but the operational definitions—not the labels—determine who belongs in each state.
Death (D): the person is dead. This state takes priority over any former care status.
Care-needing (C): the person is alive and currently meets the contract’s care trigger.
Recovery (R): the person is alive, no longer meets the care trigger, and previously met it.
Healthy (H): the person is alive, does not currently meet the care trigger, and has never met it.
Here, “healthy” is a model label, not a medical conclusion. Someone in H can have a diagnosis, use assistance, or have a limitation in one activity. The label says only that the person has not met the defined trigger and has no prior qualifying episode recorded. Similarly, R does not mean a return to perfect health: it means a former qualifying person no longer meets the trigger.
The distinction between H and R matters when future outcomes depend on history. For example, a model might need to distinguish a person who has never qualified from someone who has recovered after qualifying. If the contract treats them identically for every purpose being modelled, it may be unnecessary to keep them separate. State definitions should retain only distinctions relevant to the model’s purpose.
From Contract Wording to an Assignment Rule
A clear state definition starts with a test that can be applied consistently. In the hypothetical contract, let \(A\) be the number of activities for which inability is certified and let \(G\) indicate certified severe cognitive impairment requiring substantial supervision. Let \(L\) indicate that the qualifying activity limitations are expected to last at least 90 days. The trigger indicator is
The indicator \(Q\) equals 1 when the trigger is met and 0 otherwise. The two routes are alternatives: meeting either route is sufficient. A diagnosis that does not satisfy either route gives \(Q=0\), even if the diagnosis is serious.
A disciplined assignment order makes the categories exhaustive and prevents overlap. First check whether the person is dead. If alive, apply the current trigger. If the trigger is not met, check whether it was met previously. This order handles a person who previously qualified and still qualifies: that person is currently C, not R. It also prevents a deceased person from being classified according to a prior care state.
If the person is dead, assign D and stop.
If the living person meets either qualifying route, assign C.
If the living person does not currently qualify but qualified before, assign R.
A person who does not currently qualify and has never qualified is H.
Why the Assignment Rules Form a Partition
For a model to use these states as categories, each person must be assigned to one and only one of them. The assignment rules above guarantee this. Death is checked first, so a person cannot be assigned both D and a living state. Among living people, the current trigger divides cases into those with \(Q=1\) and those with \(Q=0\). For the latter group, prior qualification divides cases into R and H.
Let \(D\) be 1 if the person is dead and 0 otherwise; for a living person, let \(P\) be 1 if the person has previously met the trigger and 0 otherwise. State indicators can be written as
Their sum is
The bracket equals 1 because \((1-Q)P+(1-Q)(1-P)=1-Q\), and then \(Q+(1-Q)=1\). Each indicator is either 0 or 1, and exactly one is 1. This proves that the rules create an exhaustive, mutually exclusive classification, provided vital status, current qualification and prior qualification are known.
Worked Example: A Serious Diagnosis Is Not Automatically Eligibility
A hypothetical policy uses the two-route trigger defined above. Four living applicants have the following certified records: Mina has a serious neurological diagnosis but is unable to perform only one activity, and no severe cognitive impairment is certified. Oren is unable to perform two activities, with the limitations expected to last 120 days. Pia is unable to perform three activities, but the limitations are expected to last only 30 days. Quinn has certified severe cognitive impairment requiring substantial supervision and is able to perform all six physical activities.
Assumptions. Interest is irrelevant to this classification; if a rate is needed for a separate valuation, take the effective annual rate as 0%. Mortality is not modelled in these records: all four people are alive at the classification date. The records are deterministic, with no probability distribution or estimated parameters; use only the stated hypothetical contract rule and certified facts.
Method. Evaluate each route separately. The activity route requires both at least two certified limitations and an expected duration of at least 90 days. The cognitive route is satisfied only when the specified impairment and supervision requirement are certified.
Mina. Mina has \(A=1\), so \(A \geq 2\) is false. The cognitive route is also false. Thus \(Q=0\). Mina has no prior qualifying episode, so the assignment rule gives H. The diagnosis by itself does not make Mina C.
Oren. Oren has \(A=2\) and an expected duration of 120 days. Both activity-route conditions hold, so \(Q=1\) and Oren is C.
Pia. Pia has \(A=3\), but 30 days is less than 90 days. The activity route therefore fails. No cognitive route is reported, so \(Q=0\). With no prior qualifying episode, Pia is H at this date, despite having more activity limitations than Oren.
Quinn. Quinn meets the certified cognitive route, so \(Q=1\). The ability to perform all six physical activities does not negate the alternative cognitive route. Quinn is C.
Check. The result is Mina H, Oren C, Pia H and Quinn C. The classification uses the contract trigger rather than the severity of a diagnosis or the count of limitations alone. The common mistake avoided is treating “has a serious condition” or “has at least two limitations” as sufficient without checking every required condition.
Recovery Requires a History Rule
A current clinical assessment may show that a person no longer meets the care trigger, but it cannot by itself tell whether that person belongs in H or R. The distinction requires a record of prior qualification. If “recovery” is intended to represent return from a qualifying episode, its definition must say what counts as such an episode and how the model records it.
This does not require that the person be free of all impairment. A person who previously qualified because of two activity limitations and now has only one may be R under these definitions. A person who has never met the trigger and has the same current limitation is H. The difference is history, not current severity.
A model should also avoid using “recovery” to imply that the contract necessarily ends benefits or that the person can never qualify again. Those consequences depend on contract terms and model purpose. At this stage, R is simply a defined status: alive, not currently meeting the trigger, and previously qualified.
Worked Example: Distinguishing Recovery From Never Qualified
A hypothetical contract uses the same eligibility trigger. At the assessment date, all four people below are alive. Arjun previously met the certified two-activity route, but now has one qualifying limitation and no qualifying cognitive impairment. Bea has never met either route and currently has one qualifying activity limitation. Chen previously met the cognitive route and still has certified severe cognitive impairment requiring supervision. Dalia previously met the activity route, no longer meets either route, and is currently able to perform all six activities.
Assumptions. Interest is irrelevant; use an effective annual rate of 0% if a rate is required for context. Mortality is excluded because each person is stated to be alive at the assessment date. No stochastic distribution is used: these are deterministic records with the facts stated above.
Arjun. The current activity count is one, and the cognitive route is false, so \(Q=0\). Arjun has prior qualification, so the rules assign R.
Bea. Bea does not currently qualify, and has no history of qualification. The rules assign H. Bea may have a functional limitation, but that fact does not meet the stated threshold and does not establish a prior qualifying episode.
Chen. The current cognitive route is still met, so \(Q=1\). Current qualification takes priority over prior qualification: Chen is C, not R.
Dalia. Dalia currently meets neither route, but previously qualified. The rules assign R. Being able to perform all six activities is not necessary for the label R; the defining conditions are current non-qualification and prior qualification.
Check. The assignments are Arjun R, Bea H, Chen C and Dalia R. Chen illustrates why “previously qualified” alone does not determine recovery: a person who still meets the current trigger remains C. The common mistake avoided is assigning R whenever someone has any history of care qualification, without first checking the current trigger.
Qualifying for Benefits Is Not the Same as Being Paid
State definitions should specify the status they are intended to represent. If C means that the contractual care trigger is currently satisfied, it should not automatically mean that a payment is made on that date. A contract may require additional conditions before payment, such as a waiting period, claim approval or available benefit balance. Whether those conditions are represented as separate states, durations or payment rules is a modelling decision; they should not be confused with the clinical eligibility test.
Likewise, a pending assessment is not necessarily equivalent to a confirmed qualifying condition. A model may need a rule for how pending information is treated, but that rule must be stated rather than assumed. For a clean state classification, the records used should establish whether the contract trigger is met as of the assessment time.
Worked Example: State Counts From a Hypothetical Assessment File
A hypothetical insurer’s modelling team has already classified 120 people using the four rules in this tutorial. The file reports 70 living people who have never qualified and do not currently qualify, 18 living people who have never qualified but currently meet the trigger, 10 living people who previously qualified and still meet the trigger, 16 living people who previously qualified but no longer meet it, and 6 deceased people.
Assumptions. This is a deterministic summary of the assessment file, not a probability model. The implied empirical distribution assigns probability equal to count divided by 120, with category counts \(70,18,10,16,6\) and total \(n=120\). Interest is not used; take the effective annual rate to be 0% if needed. No mortality rate is estimated or applied: the six deceased records are already identified, and the remaining records are living.
Step 1: Map the file categories to states. The first group is H by definition. The second and third groups both meet the current trigger, so together they are C. The fourth group is R. The final group is D.
Step 2: Add counts by state. The care-needing count is \(18+10=28\). The complete state counts are H: 70, C: 28, R: 16 and D: 6.
Step 3: Divide by the total. The empirical proportions are
Probabilities are rounded to four decimal places. As a check, the counts sum to \(70+28+16+6=120\), and the displayed proportions sum to \(0.5833+0.2333+0.1333+0.0500=0.9999\); the difference from 1 is rounding. The common mistake avoided is counting the 10 people who previously qualified as R even though they currently meet the trigger. They belong in C under the stated priority rule.
Common State-Definition Traps
- Using diagnosis as a substitute for the trigger. A diagnosis may be clinically important but does not establish contract eligibility unless it satisfies the stated test.
- Ignoring a condition in a combined test. If eligibility requires at least two limitations expected to last 90 days, meeting the limitation count alone is insufficient.
- Calling all non-qualifying people healthy. Someone who has ceased to qualify after a qualifying episode belongs in R when the model distinguishes recovery from never-qualified status.
- Assigning recovery before checking current eligibility. A person who previously qualified but still meets the trigger is C under the definitions here.
- Equating state with payment. A person may meet the defined trigger while a separate payment condition remains unsatisfied.
- Leaving records ambiguous. State assignment requires current qualification and, when distinguishing H from R, prior qualification. If either is unknown, the classification rule needs an explicit treatment for missing information.
Check Your Understanding
For Questions 1–4, use the hypothetical trigger requiring either certified inability to perform at least two activities expected to last at least 90 days, or certified severe cognitive impairment requiring substantial supervision. Assume the stated records are complete.
- Leah is alive and has a diagnosis that limits one activity. The limitation is expected to last a year, but no qualifying cognitive impairment is certified. She has never met the trigger. Identify her state and explain which condition controls. (Answer: H; one activity limitation does not meet the two-activity threshold.)
- Marco is alive, has never previously qualified, and is certified unable to perform two activities with an expected duration of 100 days. Identify his state. (Answer: C; both conditions of the activity route are met.)
- Nia previously met the activity route. She is alive now, has one qualifying activity limitation, and does not meet the cognitive route. Identify her state and explain why the same current facts could produce a different state for someone with no prior qualification. (Answer: R; a person with no prior qualification and the same current facts would be H.)
- Omar previously met the cognitive route and still meets it. He is alive and has no physical activity limitations. Identify his state. (Answer: C; the current cognitive route is met, so current qualification takes priority over prior history.)
- A file contains 240 people: 132 who are H, 54 who are currently C and have never previously qualified, 18 who previously qualified and are still C, 24 who are R, and 12 who are D. Calculate the empirical proportion in each state to four decimal places. (Answer: H 0.5500, C 0.3000, R 0.1000, D 0.0500.)