1
00:00:04,710 --> 00:00:07,880
Let us introduce
the method we use

2
00:00:07,880 --> 00:00:12,080
for predicting
the bucket number.

3
00:00:12,080 --> 00:00:15,500
It is called-- it is a
method called classification

4
00:00:15,500 --> 00:00:16,530
and regression trees.

5
00:00:16,530 --> 00:00:19,090
In this case, we use
multi-class classification.

6
00:00:19,090 --> 00:00:22,120
There are five classes,
buckets one to five.

7
00:00:22,120 --> 00:00:24,320
To give you an example,
let us consider

8
00:00:24,320 --> 00:00:28,490
patients that have two
types of diagnosis:

9
00:00:28,490 --> 00:00:31,280
coronary artery
disease and diabetes.

10
00:00:31,280 --> 00:00:37,180
So if a patient does not have
a coronary artery disease,

11
00:00:37,180 --> 00:00:41,200
we'd classify the
patient as bucket one.

12
00:00:41,200 --> 00:00:43,480
If it has coronary
artery disease,

13
00:00:43,480 --> 00:00:46,350
then we check whether the
person has diabetes or doesn't

14
00:00:46,350 --> 00:00:47,470
have diabetes.

15
00:00:47,470 --> 00:00:53,400
If it has diabetes, then it's
bucket five, very high risk.

16
00:00:53,400 --> 00:00:55,160
And if it doesn't have
diabetes, but given

17
00:00:55,160 --> 00:00:57,380
it has coronary
artery disease, it

18
00:00:57,380 --> 00:01:01,250
is classified as bucket three.

19
00:01:01,250 --> 00:01:06,860
So this is an example in which
we only have two diagnoses

20
00:01:06,860 --> 00:01:10,970
and we will state
how the method works.

21
00:01:10,970 --> 00:01:16,880
In the application of Hawkeye,
the most important factors

22
00:01:16,880 --> 00:01:19,090
were related to cost
in the beginning.

23
00:01:19,090 --> 00:01:21,970
So in the beginning,
the classification tree

24
00:01:21,970 --> 00:01:24,680
involved divisions
based on cost.

25
00:01:24,680 --> 00:01:29,990
For example, if the patient
had paid less than $4,000--

26
00:01:29,990 --> 00:01:33,560
so this is bucket one
classification-- if it paid

27
00:01:33,560 --> 00:01:35,320
more than $4,000,
then we further

28
00:01:35,320 --> 00:01:40,250
investigate whether the
patient pays less than $40,000

29
00:01:40,250 --> 00:01:44,940
or more than $40,000
and so forth.

30
00:01:44,940 --> 00:01:48,820
As the tree grows, then
the secondary factor

31
00:01:48,820 --> 00:01:51,570
is utilized later in
the classification tree

32
00:01:51,570 --> 00:01:55,720
involve various
chronic illnesses

33
00:01:55,720 --> 00:01:58,229
and some of the medical
roles we discussed earlier.

34
00:01:58,229 --> 00:02:02,700
For example, whether
or not the patient

35
00:02:02,700 --> 00:02:05,040
has asthma and
depression or not.

36
00:02:05,040 --> 00:02:08,500
If it has asthma and depression,
then it's bucket five.

37
00:02:08,500 --> 00:02:14,740
If it doesn't, then we
consider a particular indicator

38
00:02:14,740 --> 00:02:17,440
indicating hylan
injection, which

39
00:02:17,440 --> 00:02:18,990
is an indication
of a possible knee

40
00:02:18,990 --> 00:02:20,770
replacement or arthroscopy.

41
00:02:20,770 --> 00:02:23,850
So if this indicator is equal
to 1, then it's bucket three.

42
00:02:23,850 --> 00:02:28,000
If it's indicator is equal
to 0, it's not present,

43
00:02:28,000 --> 00:02:31,660
then it's bucket one.

44
00:02:31,660 --> 00:02:35,320
So let us give some
examples of bucket five.

45
00:02:39,630 --> 00:02:44,270
So an example is as follows.

46
00:02:44,270 --> 00:02:47,680
The patient is
under 35 years old,

47
00:02:47,680 --> 00:02:53,540
he has between 3,300 and 3,900
in claims, coronary artery

48
00:02:53,540 --> 00:02:58,410
disease as a diagnosis, but no
office visits in the last year.

49
00:02:58,410 --> 00:03:00,960
Another example of a
category of a patient that

50
00:03:00,960 --> 00:03:03,680
is classified as
bucket five are claims

51
00:03:03,680 --> 00:03:09,180
between $3,900 and $43,000
with at least $8,000

52
00:03:09,180 --> 00:03:14,010
paid in the last 12 months,
$4,300 in pharmacy claims,

53
00:03:14,010 --> 00:03:18,030
and acute cost profile
in cancer diagnosis.

54
00:03:18,030 --> 00:03:21,480
And another final example is
more than $58,000 in claims,

55
00:03:21,480 --> 00:03:28,260
but at least $50,000 paid
in the last 12 months,

56
00:03:28,260 --> 00:03:31,290
but not an accurate profile.

57
00:03:31,290 --> 00:03:34,079
Classification trees
have the major advantage

58
00:03:34,079 --> 00:03:38,030
as being interpretable
by the physicians who

59
00:03:38,030 --> 00:03:40,210
observe them and judge them.

60
00:03:42,790 --> 00:03:46,970
In other words, people
were able to identify

61
00:03:46,970 --> 00:03:49,250
these cases as reasonable.

62
00:03:49,250 --> 00:03:51,590
In other words,
the human intuition

63
00:03:51,590 --> 00:03:55,990
agreed with the output
of the analytics model.