1 00:00:09,490 --> 00:00:13,120 In this lecture, we'll examine how analytics 2 00:00:13,120 --> 00:00:17,970 can model an expert, in this case a physician, 3 00:00:17,970 --> 00:00:21,730 in the context of assessing the quality of healthcare patients 4 00:00:21,730 --> 00:00:25,010 receive, and introduce a technique called 5 00:00:25,010 --> 00:00:27,750 logistic regression to achieve this objective. 6 00:00:32,369 --> 00:00:36,420 From the early 2000s, I was a member 7 00:00:36,420 --> 00:00:41,360 of the board of a company called D2Hawkeye, a medical data 8 00:00:41,360 --> 00:00:42,990 mining company. 9 00:00:42,990 --> 00:00:45,400 The company received claims data. 10 00:00:45,400 --> 00:00:47,480 These are data that are generated 11 00:00:47,480 --> 00:00:51,370 when an insured patient goes to a medical provider 12 00:00:51,370 --> 00:00:53,730 to receive a diagnosis or to have 13 00:00:53,730 --> 00:00:57,780 a procedure, for example an x-ray, or to obtain drugs. 14 00:00:57,780 --> 00:01:00,920 The medical providers need to get compensated, 15 00:01:00,920 --> 00:01:05,560 so the claims data provide the means for them to be paid. 16 00:01:05,560 --> 00:01:08,010 An important question is whether we 17 00:01:08,010 --> 00:01:13,170 can assess the quality of health care given this claims data. 18 00:01:13,170 --> 00:01:15,920 But let's first ask why assessing 19 00:01:15,920 --> 00:01:18,560 the quality of healthcare is an important objective. 20 00:01:22,789 --> 00:01:27,450 If one identifies patients that have low quality care, 21 00:01:27,450 --> 00:01:32,580 one can intervene and improve outcomes for these patients. 22 00:01:32,580 --> 00:01:37,740 Moreover, assessing quality correctly 23 00:01:37,740 --> 00:01:40,100 can control costs better. 24 00:01:40,100 --> 00:01:46,789 However, defining quality is a complex, not well-defined task. 25 00:01:46,789 --> 00:01:49,400 For example, consider what is involved 26 00:01:49,400 --> 00:01:52,470 when we talk about the quality of a book. 27 00:01:52,470 --> 00:01:58,380 It is not a well-defined, algorithmically understood task 28 00:01:58,380 --> 00:02:00,660 of defining such a quality. 29 00:02:00,660 --> 00:02:03,400 Currently, assessing quality is done 30 00:02:03,400 --> 00:02:09,070 by physicians who are experts in the health space using 31 00:02:09,070 --> 00:02:12,290 their knowledge, their expertise, and their intuition. 32 00:02:18,770 --> 00:02:23,020 So how do physicians assess quality? 33 00:02:23,020 --> 00:02:24,870 Physicians are, of course, humans 34 00:02:24,870 --> 00:02:27,310 who are limited by memory and time. 35 00:02:30,050 --> 00:02:33,730 They typically evaluate quality by examining 36 00:02:33,730 --> 00:02:36,960 a patient's records, a time consuming 37 00:02:36,960 --> 00:02:39,480 and inefficient process. 38 00:02:39,480 --> 00:02:42,970 Clearly, physicians cannot assess quality for millions 39 00:02:42,970 --> 00:02:46,160 of patients, and D2Hawkeye had, indeed, 40 00:02:46,160 --> 00:02:57,270 millions of patients who receive claims data on a monthly basis 41 00:02:57,270 --> 00:03:00,280 that the quality of them needs to be assessed. 42 00:03:04,070 --> 00:03:08,580 So the key question is as follows. 43 00:03:08,580 --> 00:03:13,480 Can we develop analytics tools that replicate expert 44 00:03:13,480 --> 00:03:17,790 assessment on a large scale? 45 00:03:17,790 --> 00:03:21,810 The goal is to learn from expert human judgment 46 00:03:21,810 --> 00:03:26,240 by developing a model, interpret the results of the model, 47 00:03:26,240 --> 00:03:31,300 and further adjust the model to improve predictability. 48 00:03:31,300 --> 00:03:34,850 The objective is to make predictions and evaluations 49 00:03:34,850 --> 00:03:41,380 on a large scale basis, to be able to process millions 50 00:03:41,380 --> 00:03:47,840 of assessing the health care quality for millions of people. 51 00:03:47,840 --> 00:03:51,100 So the lecture is a story of using analytics 52 00:03:51,100 --> 00:03:59,090 in identifying poor quality care using claims data.