|
Insurance Abstract
A method of model development for use in underwriting group life
insurance for a policy period includes collecting medical claims
data for the group to be underwritten, where each medical claim
being related to a particular employee of the group. Morbidity categories
are provided that categorize the medical claims in the medical claims
data. A conditional probability model is developed and applied to
the morbidity categories for each employee in the group using his
medical claims, thereby calculating the expected conditional probability
for each employee dying during the policy period. For each employee,
an estimate of the expected life claim cost is estimated using an
index of the life coverage to salary. Combining the expected conditional
probability for each employee dying during the policy period with
the estimate of the expected claim cost of death gives an estimate
of the group's total life exposure.
Insurance Claims
1. A method of developing a model to estimate life insurance exposure
for a group of individuals to be underwritten for a coverage period
comprising: providing medical claim data comprising a plurality
of medical claims made by a first model population having at least
a respective plurality of first model population members; providing
life insurance claim data comprising a plurality of life insurance
claims made by a second model population having at least a respective
plurality of second model population members, which optionally may
partially or wholly overlap with the first model population; providing
a clinical medical condition classification system that assigns
corresponding medical condition causes to life insurance claims
based on medical condition information included in life insurance
claims; applying the clinical medical condition classification system
to the life insurance claim data for the second model population
to generate a plurality of life insurance claims classified by medical
condition; categorizing the life insurance claims classified by
medical condition, where each morbidity category comprises one or
more medical conditions, to form a plurality of morbidity categories
of life insurance claims; calculating, for each of the morbidity
categories of life insurance claims classified by medical condition,
the proportion of life insurance claims for each morbidity category;
applying the clinical medical condition classification system to
the medical claim data for the first model population to generate
a plurality of medical claims classified by medical condition; categorizing
the medical claims classified by medical condition to form a plurality
of morbidity categories of medical claims classified by medical
condition; calculating, for each of the morbidity categories of
medical claims classified by medical condition, the proportion of
medical claims arising for each of the morbidity categories; calculating,
for each of the morbidity categories, a probability of having a
life claim during the coverage period using at least one conditional
probability, the conditional probability being based on at least
the proportion of life insurance claims and the proportion of medical
claims for each of the morbidity categories; and combining the probability
of having a life claim given a morbidity category with the expected
cost of the life claim to calculate the expected life exposure for
the morbidity category.
2. A method of estimating life insurance exposure for a group of
individuals to be underwritten for a coverage period comprising:
calculating, for each of the group of individuals, a measure of
the expected exposure comprising the application of a model developed
according to claim 1; summing, across the group of individuals and
their associated morbidity groupings, to obtain the measure of expected
future exposure for the group.
3. The method of claim 1, wherein calculating the expected dollar
amount of the life insurance claim is done by multiplying the amount
of coverage in a time period by the measure of expected probability.
4. The method of claim 2, further comprising adjusting the measure
of the expected exposure using net present valuation methods.
5. The method of claim 1, wherein each medical claim has information
about at least one medical condition, and the medical eligibility
database has information on the age and gender of each of the first
plurality of individuals.
6. The method of claim 5, wherein the life insurance claims and
medical claims are grouped into age by gender strata by each medical
condition in the clinical classification system and the age by gender
by clinical condition categories are used to define strata of expected
exposure.
7. The method of claim 5, wherein the total expected exposure is
calculated using the method of claim 1 for a block of medical claims
and it is set to equal the total expected exposure for the plurality
of individuals in the associated medical eligibility file using
standard actuarial techniques that calculate the exposure based
upon the eligible individuals' age by gender using standard actuarial
tables that provide the expected frequency of claims multiplied
by the expected exposure of that claim given the individuals' age
and gender, thereby adjusting the estimate of the expected future
exposure up or down for each clinical condition in that age by gender
stratum so the sum of the standard actuarial estimates for the individuals
in the eligibility file equals the clinically based expected future
exposure for the individuals with medical claims and were included
in the same eligibility file.
8. The method of claim 1, wherein for each individual with multiple
medical claims the are adjusted for overlapping probabilities of
death arising from the conditions indicated by the medical claims.
9. The method of claim 1, wherein the conditional probability is
calculated using Bayes Theorem or other forms of Bayesian probability
calculations.
Insurance Description
CROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application claims the benefit of U.S. Provisional
Application No. 60/701,311 filed on Jul. 21, 2005, which is incorporated
by reference in its entirety.
BACKGROUND OF THE INVENTION
[0002] 1. Field of the Invention
[0003] The present invention relates to a new technology to underwrite
group life insurance. In particular, the present invention relates
to predicting the expected claims to be made by a group based on
pooled historical data of claims for life insurance and pooled historical
data of medical claims.
[0004] 2. Background Art
[0005] Group insurance for Life Insurance (life) for employees
is widely available. In order to set appropriate premiums for these
risks it is necessary to estimate the likelihood of the insured
events (number of deaths) and the severity (or cost) of each event,
for each insured group. Because of the relative rarity of these
events for life, the experience of a group is too small to provide
reliable estimates for any but the largest groups (e.g., with ten
thousand or more employees).
[0006] The following is an example illustrating the risk profile
presented by groups of modest size, in this case for a group of
500 employees. It assumes that the binominal distribution is an
accurate representation of the likelihood of death. A 90% confidence
interval is calculated. The following table lists the range of number
of life claims in the confidence interval at 3 different probabilities
for the event. TABLE-US-00001 TABLE 1 Confidence Interval Example
90% Confidence Interval Probability Lower Bound Upper Bound .001
0 2 .005 0 5 .010 2 9
[0007] If a group has 2 events, it is in the 90% confidence interval
for probabilities .001, .005 and .010 or a potential range of a
10-fold difference in true underlying probabilities for the event.
This could result in a 10-fold difference in premium. If 0 or 1
event occurs, the underlying rate could be a .001 or .005 or a 5-fold
difference in the true probability. Therefore, refined analytic
methods are needed for accurate premium rate settings to reflect
the group's underlying risk since the 5-fold or 10-fold difference
in risk would turn into a 5-fold or 10-fold difference in insurance
premium. The 5-fold or 10-fold range in risk and premium should
be unacceptable to both the insurer and the insured.
[0008] The alternative actuarial approach uses estimates of group
risk based on the age and gender (demographics) of each group's
employees by using tables based on data pooled from many groups
(i.e., manual rates). Assume for example, the likelihood of a 20
year old male dying in a year is about 1/1,000 and the likelihood
for a 62 years old male is about 10/1,000. The group risk is calculated
by summing each eligible employee's demographic risk, the sum being
the group's base risk. The group's experience may be used to adjust
(usually done via a weighted average) the demographic risk higher
or lower, depending upon the historical experience. While the demographic
incidence rates may be modified by the industrial codes and geographic
location of specific groups they do not specifically adjust for
the considerable variation in the underlying morbidity of employees
which underlies the risks of life claims.
[0009] The experience based rates adjust for the historical or
backward looking component of underlying morbidity but do not provide
an accurate estimate of the future morbidity risk for modest size
groups.
[0010] Accordingly, there is a need for underwriting methods that
address groups of modest size and accounts for the underlying morbidity
of the employees making up a group.
BRIEF SUMMARY OF THE INVENTION
[0011] In order to improve measuring risks of death for each insured
group of employees, the present invention develops methods for achieving
the following:
[0012] 1. Providing standards for the probability of having a claim
for death based on pooled historical life claims data, that include
disease and injury specific data, measured against pooled medical
claims data.
[0013] 2. Providing expected costs or indexed cost of life claims
based on a standard unit or index value that can be scaled to the
appropriate payment amount if death occurs during the policy period.
[0014] 3. Applying 1 and 2 above to the recent medical claims data
for each insured employee group to estimate its current risks for
life.
[0015] One benefit of this invention over traditional actuarial
methods is to calculate group premium rates that are more accurate
than premium rates derived from manual, experience or combined manual
with experience based underwriting. The present invention develops
a morbidity structure and applies quantitative algorithms or statistical
models to the medical claims data for individuals in a group. The
results include predicted incidence rates of life and predicted
cost for the claim. The result of the application of this invention
is a more accurate estimate of the group's predicted exposure for
life insurance than is available through traditional methods.
[0016] Individual life insurance policies are frequently underwritten
using the results of a medical exam including hematology analysis,
urine analysis, vital signs, history and electrocardiogram. These
costly exams are used to more accurately assess and price life risk
than can be done using demographics alone. The life insurer wants
to avoid anti-selection where people with greater risk than average
apply and obtain insurance coverage without a corresponding increase
in premium. This approach is rarely used in the group market due
to its cost, intrusiveness and lack of group level benchmarks or
algorithms for aggregating results to the group level. Current individual
life insurance underwriting uses different measures of morbidity
(e.g., lab results) and different algorithms for pricing coverage.
[0017] Still further benefits and advantages of the invention will
be apparent to the skilled worker from the discussion that follows.
BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWING(S)
[0018] FIG. 1 is a flowchart of an embodiment of an overview of
a method for estimating future life claim costs and pricing the
associated insurance premium for coverage;
[0019] FIG. 2 is a flowchart of an embodiment of a process for
developing a morbidity classification system for life claims. It
details boxes 106, 108 and 109 of FIG. 1; and
[0020] FIG. 3 is a flowchart of an embodiment for applying the
morbidity classification system of cause of death to calculate incidence
rates for life claims using medical claims data for underwriting
groups and life claims databases for calibrating the life risk factors.
This details boxes 110 of FIG. 1;
[0021] FIG. 4 is a flowchart of an embodiment for estimating each
eligible person's expected life claims and combining those estimates
into a group level estimate of its claims cost and details boxes
114 and 116 in FIG. 1; and
[0022] FIG. 5 is a flowchart of an embodiment of a method for using
the expected exposure from the morbidity models to set a premium
for pricing the life insurance coverage and it details box 118 in
FIG. 1.
DETAILED DESCRIPTION OF THE INVENTION
I. Incidence Rate Calculations for Life Insurance
[0023] Although the present invention is susceptible of embodiment
in various forms, there is shown in the drawings and will hereinafter
be described a presently preferred embodiment with the understanding
that the present disclosure is to be considered an exemplification
of the invention and is not intended to limit the invention to the
specific embodiments illustrated. Accordingly, "life"
risk is used to refer to the probability of death occurring during
the policy period. "Life claim" and "death"
may be used as synonyms since a life claim results from a death.
[0024] It is to be further understood that the titles of the sections
of the specification, including but not limited to "Detailed
Description of the Invention", relate to a requirement of the
United States Patent and Trademark Office, and are not intended
to, do not imply, nor should be inferred to limit the subject matter
disclosed herein or the scope of the invention.
[0025] Referring to FIG. 1, an exemplary embodiment of the present
invention collects life claim information (box 102) and also collects
medical claims information (box 104). A clinical classification
system is developed for the cause of the life claims in the database
(box 106) and applied consistently to both the life claims and the
medical claim data (box 108). The clinical classification system
is then collapsed into a more tractable number of categories, called
morbidity categories, in order to get more stable incidence rate
estimates (box 109). Models for the incidence of life claims are
calibrated (box 110). The medical claim and eligibility databases
are then scored (box 114). The results of that scoring can then
be used to estimate the number and cost of expected claims which
can be used to estimate total expected claims costs or life exposure
(box 116). The results of the scoring can then be used to determine
pricing for life coverage (box 118).
[0026] This invention uses an alternative approach to traditional
experience or demographic based underwriting. This invention can
use life claims data (particularly diagnostic information--see box
102 in FIG. 1 regarding life claims data and below for exemplary
data layout in Table 2) as a first database or life claims database.
TABLE-US-00002 TABLE 2 Sample Life Data Layout Variable Type Format
sex Char $4.00 Dx1 principal Char $6.00 cause death Dx2 secondary
Char $6.00 cause death Dx3 tertiary Char $6.00 cause death patdob
Num YYYYMMDD Death date Num YYYYMMDD Reported date Num YYYYMMDD
Paid date Num YYYYMMDD age Num 3 male Num 1
[0027] Where sex is the gender of the subject, Dx1 (ICD-9 code)
is the diagnosis that was the principal cause of death, Dx2 is the
diagnosis that was the secondary cause of death, Dx3 is the diagnosis
that was the tertiary cause of death, patdob is the claimant's date
of birth, Death date is the date of the death, Reported date is
the date the death was first reported, Paid date is the date the
death claim was paid, age is the age of the claimant and male is
1 if male claimant and 0 otherwise.
[0028] Medical claims data must include computerized diagnostic
information to augment demographic data--see box 104 of FIG. 1 and
below for exemplary data layout in Table 3 (or medical claims database)
and Table 4 is a second database (or medical enrollment database).
TABLE-US-00003 TABLE 3 Samples Medical Claims Variables (Employees
Only) Variable Type Format group Char $14.00 personid Char $20.00
DX1 Char $6.00 DX2 Char $6.00 DX3 Char $6.00 DX4 Char $6.00 DX5
Char $6.00 Paid date Num YYYYMMDD Incurred date Num YYYYMMDD
[0029] Where group is the employer ID, personid is the employee's
ID, DX1-DX5 are up to 5 ICD-9 code diagnoses with the decimal that
are associated with that employee's claim, Paid date is the date
the claim was paid and Incurred date is the date that the service
was incurred. Appendix 1 contains an example SAS data standardization
program for a typical medical claims database that includes charges,
payments and other variables not required for this invention but
are typically included in medical claims data files. TABLE-US-00004
TABLE 4 Samples Enrollment Variables (Employees Only) Variable Type
Format group Char $14.00 SEX Char $1.00 personid Char $20.00 PATDOB
Num YYYYMMDD
Where group is the employer, SEX is the gender of the employee,
personid is the employee's ID number and PATDOB is the employee's
date of birth. Note that dependents may need to be screened from
the eligibility file since they are typically not covered under
group life policies.
[0030] The present invention does not require that these databases
be linked at the person level but nevertheless may be practiced
with some or all of the data being linked, though with all the data
linked other methods would probably be used by those of ordinary
skill in the art. A practicable exception is for the group and employee
linkage between the medical claims and eligibility file as per the
data in Tables 3 and 4 above, but with no linkage between Table
2 and Table 3 or 4. In other words, a large life claims database
can be used with a large medical claims database possibly from a
separate population to develop incidence rate models for life claims.
[0031] To facilitate the collection of medical data as exemplified
by box 104, below is a sample program to standardize life medical
claims data with ICD-9-CM (or 10) codes (the standard for U.S. medical,
life and disability insurers for coding the cause of a claim, and
these data can be 5 digit or 3 digit ICD codes). Other coding or
medical classification systems also exist, e.g., SNOMED. For the
purposes of the present invention codes like ICD-9-CM or ICD-10,
SNOMED, or other systems that are used to categorize medical conditions
are referred to as medical condition codes or medical diagnosis
codes or, for shorthand, medical diagnoses.
[0032] The data can be read from an Excel spreadsheet, for instance,
to be analyzed in a computer system coded to implement the present
invention. One system that can be used to implement the present
invention is the commercially available statistics program SAS produced
by SAS Institute of Cary, N.C. The input data are converted to the
format needed (i.e., standardized) by the analysis system and variable
names from the life claims database are mapped to standardized names
utilized by the analysis system. In the present examples, the medical
condition codes used by the analysis system are modified ICD-9-CM
codes, but other code systems or modifications of the ICD-9-CM system
may be used.
[0033] Data can be validated by comparing summarized data from
the medical claims and life claims to control totals (e.g., total
number of records, number of claims, total claims paid, etc.) and
other summary reports developed by the organization supplying the
data--standard computer processing techniques for validating accurate
data transfer and reading. The data can also be standardized into
common units such as date format, level of medical condition code
used (e.g., for ICD-9, 5 digit, 3 digit, subchapter grouping) and
currency units.
[0034] A universe of claims eligible for analysis and use in the
database can be defined and used to create and calibrate the life
prediction models. A universe of eligible claims can exemplarily
comprise data such as date range for claims, the covered person's
age and gender, and the covered person's employment status. When
each source is complete the files can be concatenated into one file.
[0035] Continuing the discussion of box 104 in FIG. 1, in Appendix
1 is a SAS program for medical claims data to standardize its format
for processing. Each source of data may have a different layout.
The medical claims data can be validated, cleaned and then standardized.
[0036] A universe is defined and data that have similar formats
from different sources can be concatenated into one file for summarization
and analysis.
[0037] The coding of the medical claims and life claims with medical
condition codes should be at the same level of specificity, and
as indicated earlier, is not limited as to form. The medical condition
coding system can have different levels of specificity, and may
be hierarchical in nature such that a higher level of a hierarchy
may subsume multiple elements of a more detailed hierarchy below.
Where medical claims data and life data are not in a consistent
format, but can be mapped onto the same system, where the system
mapped to is hierarchical the coarsest level may need to be used
for comparability across data sets. Additionally, common but inconsequential
diagnoses will be ignored in the medical claims when the life claims
data do not contain that level of detail, that is to say that while
a condition may be categorized for medical care purposes, it has
no presence in databases tracking causes of life.
[0038] The medical claims and life databases are usually from different
populations. The resulting morbidity based life incidence models
are then applied to different groups' medical claims data to produce
expected life exposure for underwriting life insurance. This refined
estimate of claim frequency and severity can then be used in estimating
the costs associated with insurance coverage and accordingly pricing
the insurance coverage.
[0039] The morbidity model based insurance premium pricing can
be an adjustment of the traditional demographic or experience based
methods used by actuaries and underwriters or derived separately
using exposure data and actual claims experience. Life cost and
total exposure estimates derived from the medical claims data for
pricing life will be described later.
[0040] Often, the age by gender specific incidence of medical conditions
or morbidity can be similar in the insured populations of employees
for both the medical claims database and the life claims database,
and incidence rate estimation can be done on that basis.
[0041] Large life and medical claim files help assure that they
are more likely to have similar morbidity patterns (within age by
gender strata) than smaller, more selective files.
[0042] Use of standard Society of Actuaries (SOA) age by gender
standards to estimate life claim risk also involves a similar operating
limitation; specifically, that the morbidity pattern in the population
used as the SOA standard is similar to the life insured population
being priced. This is because morbidity of the insured population
drives the system, not age or gender of the insured population directly.
[0043] Under the conditions of similar proportions of medical conditions
(within age by gender) for the medical population and the life population,
the relative incidence rate can be calculated for such conditions
using conditional probabilities or similar measures. Traditionally,
the incidence rate for life is the number of life claims in a year
divided by the number of insured people and stratified by combinations
of age, gender, occupation and other factors related to life claims.
However, these calculations of incidence rates are not done by morbidity
condition with current underwriting procedures for groups.
[0044] This method and system of the present invention calculates
the incidence rates conditionalized on morbidity condition can be
calculated in addition to demographics and other factors. While
any conditional probability estimation method can be adapted for
the practice of this invention, Bayes Theorem is used for the preferred
embodiment of this invention: Probability .times. .times. ( A .times.
.times. given .times. .times. B ) = probability .times. .times.
( B .times. .times. given .times. .times. A ) .times. probability
.times. .times. ( A ) probability .times. .times. ( B )
[0045] This calculation is normally done within an age by gender
strata, but the calculation can also be applied to data having morbidity
information that has not been stratified. A concrete example is
provided below. For life insurance:
[0046] Probability (death given diagnosis Y)=probability (diagnosis
Y given death)* probability (death)/probability (diagnosis Y), where
the probability (diagnosis Y given death) equals the proportion
of life claims resulting from diagnosis Y (a morbidity category),
the probability of diagnosis Y equals the proportion of the insured
population with diagnosis Y, and the probability of death is the
death rate for the age by gender strata for a large insured population
with similar characteristics.
[0047] The following table is a hypothetical example of a Bayesian
calculation for the probability of death for a male 55-64 years
old with lung cancer. Assume the following: lung cancer is 5.0%
of death claims for males 55-64 (=probability (B given A) from equation
above); the probability of death is .007 for 55-64 males during
the next year (=probability (A) or the prior from equation above)
and the probability of lung cancer for males 55-64 is 0.6% (=probability
(B) from equation above). This results in a Bayesian probability
estimate of 0.058335 (=probability (A given B) from equation above)
or the probability of a life claim given lung cancer is 5.8% using
these calculations, as shown in Table 5. TABLE-US-00005 TABLE 5
Male 55-64 Death Medical Lung cancer count 60 70 Total count 1200
11667 % 5.00% 0.60% Prior probability of 0.70% death male 55-64
Bayesian probability = (.05 * .007)/.006 Bayesian probability (death
given lung cancer) = 0.058335
[0048] Therefore, the probability (death given diagnosis Y) can
be calculated using readily available medical and life insurance
claims data. This is the fundamental building block for calculating
morbidity based death rates.
[0049] A more general alternative to Bayes Theorem for calculating
conditional probabilities is a Bayesian Network where Bayes. Theorem
is applied multiple times in the context of a decision tree, conditionalizing
the probability on variables such as age, gender, standard industry
classification code (SIC code) of employer, occupation, geographic
region, company size and experience and other factors in addition
to the morbidity. Graph theory (especially directed or acyclical
graphs), Monte Carlo simulations and other techniques can be used
to calculate the joint and conditional probabilities needed for
predicting the incidence rate of life claims for individuals and
groups of employees. The result of these joint and conditional probability
calculations is a look-up table of incidence rates that is used
to score each person in a group. The person's characteristics are
classification factors for the table and the associated incidence
rate for life is a predicted conditional probability that is looked
up in the appropriate cells of the table.
[0050] Referring to box 106 of FIG. 1, the preferred embodiment
uses groupings of ICD 9-CM (or 10 or other coding schemes) diagnosis
codes to define morbidity categories. The coding system should be
the same for medical and life claims or be able to be mapped uniquely
onto a common coding system. The exemplary embodiment of the present
invention uses ICD-9-CM, the present industry standard coding system.
However, the present modeling approach can be applied to other classification
systems with just as much validity. The accuracy of the present
modeling approach will be dependent upon the accuracy of the classification
system, the coding accuracy, the representativeness of the sampled
population and the amount of data available for model calibration.
The morbidity categories should be homogeneous (see boxes 204 and
206) with similar likelihood of leading to death. Non homogeneous
categories can be used if the likelihood of death is similar by
medical condition code within the category. The objective is to
develop morbidity categories that are clinically similar or at least
have similar risk (e.g., Bayesian probability of the insured event)
and are large enough to lead to robust estimates of the incidence.
[0051] A life insurer typically does not have medical information
on a large portion of its insured customers so the conditional and
joint probabilities cannot be calculated directly. This invention's
preferred embodiment calculates the probability (death given diagnosis
Y) directly using the definition of a conditional probability when
appropriate data are available (medical and life claims are linked
at the person level) and using a technique such as Bayes Theorem
or Bayesian Networks when the data do not support a direct calculation
of the conditional probability. A simple example has 10,000 males
55-64 covered for medical and life. 6 of those males covered have
life claims from lung cancer and a total of 100 have lung cancer
as a medical condition, resulting in a conditional probability of
6/100 or 6% of having a life claim given that you are a male 55-64
and have lung cancer.
[0052] Referring to box 109 in FIG. 1, the medical and life claims
are classified by morbidity category. The following is one simple
example describing using a listing of ICD-9 codes as one possible
definition of the lung cancer morbidity category. Other definitions
may be used also or the category combined with other cancers, etc.
The number is the ICD-9-CM diagnosis code and the following text
is an abbreviated name (MAL or MALIG "malignant", NEO
means "neoplasm", NEC means "Not Elsewhere Classified",
and NOS means "Not Otherwise Specified") of the site of
the lung cancer. [0053] 162 MAL NEO TRACHEA/LUNG* [0054] 1622 MALIG
NEO MAIN BRONCHUS [0055] 1623 MAL NEO UPPER LOBE LUNG [0056] 1625
MAL NEO LOWER LOBE LUNG [0057] 1628 MAL NEO BRONCH/LUNG NEC [0058]
1629 MAL NEO BRONCH/LUNG NOS [0059] 164 MAL NEO THYMUS/MEDIASTIN*
[0060] The calculation using Bayes Theorem of the probability (death
given diagnosis Y) equals the (proportion of the life claims with
diagnosis Y as the cause--calculated from the life claims file)
times the (overall incidence of death) divided by the (probability
of diagnosis Y) which equals the proportion of people with diagnosis
Y in the medical claims database.
[0061] Referring to box 110 in FIG. 1, the proportion of people
with diagnosis Y is calculated as the number of people with a medical
claim of diagnosis Y divided by the total number of people enrolled
for medical coverage. When the medical claims and life claims are
directly linked by a unique person level identifier (from the same
person the conditional probability can be calculated directly using
the formula
[0062] Probability (death given diagnosis X)=Probability (death
and diagnosis X)/ Probability (diagnosis X)
[0063] This is the preferred embodiment when the data are linked
at the person level although such data are rare, and the invention
can be used where such data is not linked at the person level.
[0064] Calculations are typically done using age ranges (frequently
9 ranges of 5 years, such as 20-24 (or 18-24), 25-29, . . . , 60-64
by male/female) for groups of insured employees. This process makes
an array of boxes, one axis being diagnosis, and the other an age
range (see Table 6 for an example). TABLE-US-00006 TABLE 6 General
Category for Incidence Rate Calculations Age 18-24 Age 25-29 Age
30-34 Age 35-39 Infectious 5 52 33 10 diseases
[0065] Estimates of the Bayesian or conditional probability of
death given a diagnosis X are calculated for a set of morbidity
conditions within each age by gender strata. The number of morbidity
conditions that are useful will be driven by the number of life
claims by strata and the frequency by morbidity conditions. Fewer
than 5 claims per age by gender strata or cell is not very credible
for probability calculations. At least 10 claims per cell are recommended;
however, adjacent ages or gender can be combined for greater counts,
providing greater rate stability if they have similar underlying
distributions. For example, sometimes you merge two or more adjacent
but not credible boxes to make a contiguous credible box. Table
7 shows two generic cases, and Table 8 shows a concrete example
of merging categories from Table 6. TABLE-US-00007 TABLE 7 Age 1
Age 2 Age 3 Age 4 male Merged gender female Merged adjacent age
example example
[0066] TABLE-US-00008 TABLE 8 General Category for Incidence Age
18-24 Age 30-34 Rate Calculations and Age 25-29 and Age 35-39 Infectious
diseases 57 43
[0067] The enrolled medical population with medical claims typically
will be larger than the life claims file since the frequency of
employees incurring one or more medical claims in a year is about
600-800/1,000 enrollees per year versus 1-10/1,000 enrollees for
death per year. Therefore, definitions of morbidity conditions producing
credible probabilities within the life files usually provides credible
probability estimates in the medical claims calculations since the
higher frequency of medical claims than life claims gives larger
number of cases and more credible estimates of incidence rates.
[0068] Box 206 of FIG. 2 and the table for Males 55-64 with EP
180, and lung cancer illustrate Bayesian probability calculations.
This example is based on disability rates but the application to
death rates is obvious to one skilled in this art. When the incidence
rates of death are different for similar conditions, those conditions
should be put into separate morbidity categories for predicting
future life claims for underwriting (see box 208).
[0069] Many life claim files have a single medical condition code
as a cause of death. People in medical plans may have numerous diagnoses
in a year. (Note that one year is used in this invention as the
standard contract time period for pricing and rate estimation even
if the insurance contract is for a different time period since one
year eliminates seasonality and most claim files have observations
based on groups enrolled for multiples of a year but other time
periods can be used by rescaling all of the calculations.) If the
life claims file has multiple codes, then morbidity categories can
be defined as combinations of multiple medical condition codes.
Otherwise, there will be multiple morbidity categories associated
with individuals in the medical claim files and only one morbidity
category associated with the life claim.
[0070] This potential multiple counting of individuals or morbidity
categories can be adjusted for in many ways. The most straightforward
adjustment is to assume the highest probability morbidity condition
only for an individual. One alternative is that all morbidity conditions
can be used and assume statistical independence between the conditions.
The overlap between the pairs, triples, etc. of morbidity conditions
needs to be calculated so that they are not double counted. The
following is a general formula for calculating the union of independent
events A1 through An. P(.orgate. Ai for i=1, . . . ,n)=.SIGMA. P(Ai)-.SIGMA.ijP(Ai,
.andgate.Aj)+.SIGMA.ijk P(Ai, . . . .andgate. A.j..andgate. A .
. . k)-. . . +(-1).sup.n-1P(.andgate. Ai for i=1, . . . , n)
[0071] If the morbidity conditions have a positive interaction
with one another (the likelihood of death is greater than the sum
of the parts) then double counting would be a more accurate estimate
of the probability of a life claim than the union of the probability
assuming statistical independence.
[0072] The preferred embodiment of this invention normalizes or
sets the estimate of the total number of expected life claims for
a large block of business (e.g., all groups in a large medical plan)
to the expected number of claims for that block calculated using
a traditional actuarial model driven by demographics and other non-morbidity
factors. Morbidity categories with few or no life claims can be
omitted. The method used for adjusting the multiple morbidity conditions
in the medical claims will not have any influence on the total expected
number of life claims for the block of business if the block is
set to equal the number of expected claims from the demographic
model. This assumption is not necessary but can be made. However,
there could be large differences at the person and group levels.
[0073] The medical condition codes for a person with medical claims
are combined into a non-duplicated array covering a fixed time period
such as a year (see box 302). Each distinct medical condition code
is then classified into a morbidity category and duplicates of each
category are dropped (see box 304). All people in the eligibility
file, including those with and without medical claims, are also
classified into age by gender strata regardless of whether they
have a medical condition code (see box 306). External age by gender
rates of death (see box 308) may be used to calibrate the overall
incidence rate estimates to established standards. Those rate estimates
are used in the numerator of the Bayesian probability calculations
as the prior probability of death (see box 310).
[0074] A normalization of the incidence rates for life insurance
(see 312) is done by age by gender strata across all morbidity categories
by setting the total estimate for the morbidity model for each demographic
stratum for a block of business to equal the total estimate from
a demographic model for the same demographic strata. In other words,
the normalization is a simple, proportionate adjustment increasing
or decreasing the morbidity model prediction so that the number
of expected life claims from the morbidity model equals the expected
number of life claims from the demographic model for a large book
of business. Dividing the demographic estimate by the morbidity
estimate produces the normalizing factor that will equalize the
morbidity and demographic model estimates. This factor is then multiplied
by each morbidity incidence rate estimate. This equalizes the morbidity
and demographic incidence rates for the entire block of business
by demographic strata but allows individual group estimates to vary
substantially. The strata and categories may be very fine or to
only a few rates, depending upon the accuracy and amount of data.
This normalization is not necessary but optional.
Life Insurance Claim Cost
[0075] Life insurance typically consists of a single payment that
is usually a multiple of the employee's salary. If the proportion
of life claims by morbidity category varies substantially by salary,
then multiple sets of conditional probabilities may need to be calibrated
on life claims data stratified by salary level. Otherwise, an index
value, such as the multiple of salary that is the life insurance
coverage, can be used for the cost of the life insurance claim.
The index value is then scaled by the salary or other factor that
will translate the index amount into the life insurance liability
in dollars. If life claims are not paid quickly, discounting may
be used to provide a more accurate estimate of the life insurance
liability to the insurer. Discounting is performed using standard
actuarial techniques for discounting future payments.
[0076] The discounted expected claim cost and the incidence rate
from the Bayesian probability model are calculated for each separate
morbidity category for each person (see box 313). The expected cost
is multiplied by the Bayesian probability to calculate the total
expected discounted cost for that morbidity category for that person.
Each person's expected discounted exposure by morbidity category
(product of probability times cost) is summed to calculate that
person's total expected discounted exposure for the preferred embodiment.
The objective is to provide the most accurate estimate of life claims
cost for the group for use in setting the premium.
[0077] Calculating Person And Group Level Expected Life Claims
Costs
[0078] Referring to boxes 114 and 116 of FIG. 1, in order to price
the insurance, an expected claim cost or life exposure must be calculated.
Estimating the probability of a person having a life claim is done
by "scoring" the medical claims data (see also FIG. 4).
A person level file is prepared that includes the person's age,
gender and morbidity grouping(s) from the medical claims. The preferred
embodiment uses a look-up table to associate the morbidity grouping(s)
by age by gender strata (alternatively other factors such as company
SIC code, geographic region and others can be applied) with the
probability of death (see box 402). Multiple look-up tables may
be required if alternative insurance options are required for pricing
the life coverage. For example, the occupation may have a large
impact on the probability of death. Therefore, different incidence
rates by demographic strata should be used in the Bayesian probability
calculation if sufficient data are available for accurate rate estimates
and scoring by occupation. If death rates by occupation are not
credible, then occupational categories need to be collapsed to produce
stable death rate estimates.
[0079] Each person with morbidity grouping codes found associated
with death is scored by looking up the probability of death associated
with their demographic stratum and each of their morbidity grouping
codes associated with death. Morbidity codes that are not associated
with death in the life claim file may be assigned zero probabilities.
Alternatively, the likelihood of death by accident, homicide or
suicide can be calculated and allocated according to demographics
of all eligible employees, replacing the zero incidence rates for
people with no medical claims. If this approach is used, those morbidity
categories should be removed from the conditional probability calculations
to avoid double counting.
[0080] The likelihood (probability or incidence rate from box 402)
for a morbidity grouping must be multiplied by the mean of the expected
discounted cost (assuming the death occurs) to calculate the expected
value of the discounted cost for a person with a morbidity condition
(see box 404). This is the expected value in expected discounted
cost for that person for that morbidity grouping. The preferred
embodiment assumes a zero probability of a death claim from people
with no diagnosis in the medical claims data other than from accident,
homicide or suicide approach described above. Alternatively, the
likelihood of sudden death from non-diagnosed morbidity may be estimated
and included in the probability calculations. However, care must
be taken to avoid double counting so renormalization of overall
death rates to a large block of business may be required.
[0081] The preferred embodiment uses double counting of the probability
overlaps when a person has multiple morbidity groupings-assuming
that the events are additive but that the sum can be no greater
than 1.0 (a certain event). The largest sum of probabilities for
an employee is likely to be far less than 1.0. The expected claim
cost is multiplied by the sum of the probabilities of his or her
morbidity conditions leading to death in the next year. This estimate
includes double counting for people with multiple morbidity conditions,
as previously discussed. Therefore, the sum of all individuals'
expected claim cost within a demographic stratum (e.g., males 18-24)
for a large block of business may require recalibration to eliminate
the double counting. One can normalize the morbidity model by multiplying
each person's estimate by the ratio of the demographic stratum's
estimate divided by the morbidity estimate for those within that
demographic stratum, producing total expected morbidity claims costs
equal to the entire block for that demographic stratum (see box
402). The ratio of an individual group's (e.g., an employer) demographic
based expected claims cost divided by the morbidity model's expected
claims cost is a measure of the relative morbidity risk for that
group.
[0082] Actual underwriting results should be tracked over time
and compared to the expected results. This information can be used
to modify the incidence and cost models, especially for eliminating
any systematic biases where the predictions are too high or too
low (see box 116 in FIG. 1 and box 414 in FIG. 4).
Indication of a Price Quote and Morbidity Profile
[0083] The medical claims data and medical eligibility may not
be sufficient to provide a final quote for life coverage since those
files may not have an accurate census for life, indicate the amount
of coverage such as average wages, show the group's life experience,
or provide the industry type and occupation. However, a meaningful
indication of the approximate proportion of the final quote to the
current rates can be estimated by dividing the morbidity based estimate
of claims costs by the demographic based estimate of claims costs
(morbidity to demographic ratio).
[0084] This invention has an option of sending the morbidity to
demographic ratio to the employer or broker representing the group.
This indication would be sent to groups with morbidity to demographic
ratio meaningfully below 1.0 along with a request for the information
necessary to provide a final, binding life insurance quote.
[0085] In addition, a morbidity profile may be provided that compares
the employer to a norm by morbidity category. The morbidity profile
compares the percentage of employees with a morbidity condition
that was used in the morbidity prediction model. The norm may be
age by gender adjusted so that the employees' age or gender skew
does not drive the differences. The objective is to provide information
on morbidity conditions where the group may have an excess of morbidity
so that disease management or other programs may be used to help
reduce morbidity levels, thereby lowering the group's employees'
risk of death. Morbidity improvement will lower future insurance
costs and the employer's cost of turnover. Note that the conditions
profiled are those morbidity conditions that are related to death
rather than future medical expenses. Profiles of a group's frequency
of actual morbidity groupings from the medical claims versus an
average or standard may be provided to better understand the potential
cost of death for a group or block of business.
Pricing Groups
[0086] Current actuarial estimates of risk frequently involve blending
or averaging manual rates with experience based rates. Credibility
theory is used to determine the weights assigned to each estimate.
This concept can be generalized to include a morbidity based estimate
of cost. The benefit of this approach is that the morbidity information
is from details of the group's potential exposure while the manual
rate reflects population averages and the experience provides information
on the group's historical death rates. Weighting and combing these
factors provides a unique estimate that should be more robust than
estimates derived without the morbidity model's expectation.
[0087] The morbidity to demographic ratio (i.e., morbidity model
based expected claims cost divided by demographic model based expected
claims cost) can be used to adjust the traditional premium to more
accurately reflect the morbidity characteristics of the group and
its associated claims costs (see box 118 in FIG. 1 and box 502 in
FIG. 5). When the morbidity to demographic ratio is meaningfully
greater than 1.0 (see box 508), the demographic characteristics
do not entirely account for the group's morbidity risk so the premium
should be raised. The traditional premium can be multiplied by the
morbidity to demographic ratio. Alternatively, a weighted average
of the two methods may be taken based on the credibility of the
two methods used or another method for combining the two estimates
may be used for estimating expected claim costs. The preferred embodiment
gives 100% credibility to the morbidity to demographic ratio so
the traditional premium is multiplied by the morbidity to demographic
ratio. This will discourage groups from accepting coverage without
paying additional premium to compensate for its morbidity risk.
[0088] Conversely, when the morbidity to demographic ratio is meaningfully
less than 1.0 (see box 504), its morbidity risk is lower than its
demographic risk. The ratio can be multiplied by the traditional
premium estimate to produce the claim cost estimate. However, a
smaller decrease may be enough price inducement to compel the customer
to select this coverage over traditionally derived (and more expensive)
premiums. For example, only half of the discount may be applied
to the premium. Generally, the offered premium should be between
the two estimates and this is the preferred embodiment.
[0089] Typically, many groups will have morbidity to demographic
ratio that is close to 1.0 (see box 506). Those groups can be priced
at the demographic based premium (manual rate) or a combination
of the manual rate multiplied by the morbidity to demographic ratio
or some other combination of the two factors.
[0090] Experience may be blended with the morbidity to demographic
ratio. However, the confidence interval is frequently large so random
variation may can be added to the cost estimate using experience.
The preferred embodiment uses the experience to modify the morbidity
to demographic ratio when the experience is worse than the upper
bound of a large confidence interval (e.g., 90%).
[0091] An estimate of the variance of the group may be calculated
as an additional pricing factor with greater variance implying greater
risk and a higher price with all other factors being equal. The
Lexian probability distribution can be used to calculate a confidence
interval about the group or block's average expected death rate.
[0092] The Lexian distribution is somewhat more accurate than the
binomial or Poisson distribution which could be used as alternative
variance estimates. Larger confidence intervals for the expected
claims cost imply a higher premium, all other factors being equal.
[0093] An alternative approach uses Monte Carlo simulation for
estimating life pricing. The Monte Carlo simulation will calculate
numerous random samples for a group using that group's morbidity
categories and demographic stratification. The incidence rate for
an individual will be 0 or 1, selected randomly in proportion to
the morbidity incidence for that person's morbidity category by
age by gender strata. If 1 is selected, then the expected claims
cost is added to the groups sum of total expected claims costs.
When zero is selected, zero expected claims cost is added to the
sum of the total estimates for that group. The expected claims costs
are summed for all people in that group. This process is repeated
with random selections numerous times providing a distribution of
expected claims cost for the group. This is the preferred embodiment
for calculating the distribution of total life claims cost for a
group or block of business.
[0094] The final step in the product is offering the insurance
coverage to the group or its agent at the modified premium (see
box 510). Results (e.g., loss ratios, acceptance of the insurance
coverage) should be tracked and can be used to modify the algorithms
for future policy offerings and premium rate setting (see box 512).
[0095] From the foregoing, it will be observed that numerous modifications
and variations can be effectuated without departing from the true
spirit and scope of the novel concepts of the present invention.
It is to be understood that no limitation with respect to the specific
embodiment illustrated is intended or should be inferred. The disclosure
is intended to cover by the appended claims all such modifications
as fall within the scope of the claims.
[0096] Each of the patents and articles cited herein is incorporated
by reference as if fully set forth herein. The use of the article
"a" or "an" is intended to include one or more.
TABLE-US-00009 APPENDIX 1 A sample SAS read program for medical
claims using TruRisk's standard layout is below (see Box 104). Each
client may have a different layout so the read programs will vary.
The medical data are validated and then standardized. A universe
is defined and data from different sources are concatenated into
one file for analytics. A sample read and standardization for one
source follows. libname dat `.`; %let tpa = SOURCE; run; filename
filein ("medical080204.ic.clm"); run; title "&tpa
test claims"; run; data dat.claims; infile filein lrecl=400
pad; length group $14. EmpID $11. depid $2. clmno $10.; input @1
Group $11. @12 EmpID $18. @30 DepID $3. @33 Clmno $19. @52 Clmline
$5. @57 servDate mmddyy8. @67 paidDate mmddyy8. @77 Charge 17. @94
Allow 17. @111 Deduct 17. @128 Copay 17. @145 Coins 17. @162 Payment
17. @179 PPOSave 17. @196 dx1 $5. @201 dx2 $5. @206 dx3 $5. @211
dx4 $5. @216 dx5 $5. @221 CPT $5. @226 HCPCS $5. @231 AdjFlg $2.
@233 OrgClmNum $19. @252 OrgChrgNum $19. @271 pos $5. @276 notALLOW
17. @294 COB 17. ; *Create unique identifier for a person for each
group; length personid $14. ; depid=translate(right(depid),`0`,``);
personid=compress(left(EmpID)||left(DepID)) ; if compress(clmtype)
eq :**<<< select only medical; format paiddate servdate
mmddyy8.; run; **validate and standardize; data dat.claims dat.baddates
dat.badempids; set dat.claims; drop notallow pposave deduct ; allow=charge-notallow;
inelig=notallow; discount=pposave; deduct_c=sum(deduct,copay); coins=.;
inpat=0;outpat=0;home=0;snf=0;urgent=0;er=0;ambsurg=0; if compress(pos)
in (`1`,`01`,`21`,`IH`,`I`) then inpat=1; if compress(pos) in (`2`,`02`,`22`,`OH`)
then outpat=1; if compress(pos) in (`4`,`04`,`12`,`H`) then home=1;
if compress(pos) in (`8`,`08`,`31`,`SNF`) then SNF=1; if compress(pos)
in (`20`) then URGENT=1; if compress(pos) in (`23`,`ER`,`E`) then
ER=1; if compress(pos) in (`24`,`ASC`,`B`) then AMBSURG=1; if servdate
eq . or paiddate eq . then output dat.baddates; else if empid eq
` ` or empid eq :`UNEXP` then output dat.badempids; else output
dat.claims ;
|