Provider Continuity in Taiwan 台灣醫療照護持續性 蒲正筠 18.03.2019 國立中正大學

蒲正筠 Christy Pu Institutes Degree Department Period National Yang-Ming University (Taiwan) PhD Public Health 09/2005~06/2008 University of Oxford (UK) MSc Economics 08/2002~07/2003 University of the West Indies (Trinidad) BSc 08/1999~07/2002 現職 國立陽明大學公共衛生研究所 教授 國立陽明大學 副學務長 國立陽明大學公共衛生研究所政策法律組召集人 研究興趣 病患行為 醫療費用與自費醫療 醫療會計帳

何謂照護持續性? Continuity of Care COC (照護持續性) has many definitions/aspects. 泛指病患和醫生之間的關係，而非針對特定疾病。 ‘Continuous caring relationship’ with an identified health care professional. Can also mean “health care is provided for a person in a coordinated manner”. 分為三大類: (1) 資訊持續 Information continuity, (2) 管理持續 Management continuity, (3) 關係持續 Relational/provider continuity

COC在有「基層醫療」的國家是什麼意思? 病患接受到協調性且為間斷的醫療服務 有明確轉診制度，所以初級轉次級會(要)有好的協調 更好的醫病關係 病歷(病史)不中斷 、不會開錯藥 醫師因為被病患信賴，所以更加仔細、更有責任感….

COC在「台灣」是什麼意思? 台灣是以專科醫療照護為主的國家 沒有完整的轉診制度、沒有「家庭醫生」 雲端藥歷、院所共享病歷 不同疾病本就是看不同專科醫師 全民健保造成亂逛醫院，因此「亂換醫生」確實是重要議題。 隨著big data普及，國際間發展出可量化之照護時續性指標，但是只有「關係持續」(relational continuity)的指標。

如何測量 relational (provider) continuity? the COC index (COCI): Where N is the total number of visits, nj is the number of times the patient visited a physician j; M is the total number of physicians visited. The index value ranged from 0 to 1, where 0 indicated no continuity and 1 indicated perfect continuity. Why better patient outcome? Better trust, the physician knows your family history and medical history, better communication, clinical responsibility.

無法反映 這類COC量化指標有什麼限制? 呈現 研究考量 照護持續性指標僅看得出有沒有換醫生，看不出有沒有「亂」換醫生。 換醫生不代表病患沒有受到「協調性」照護。 呈現 無法反映疾病特質、病患特質 (例如CCI共病指數) 無法反映不換醫生是因為有良好關係或是其他因素 (例如當地只有那位醫生) 無法反映 研究上也需要考量reverse causality (因為治不好所以換醫生) 其他confounder: 供給還是需求因素? 研究考量

健保資料庫 Claims data/Administrative data 兩百萬歸人檔 全人口檔 所有健保給付的醫療服務 就醫日期、費用、疾病診斷、用藥、手術、 醫院、醫師…等 病患承保資料、醫院資料 ICD 9/10 (2000-2015年採ICD9；2016年後採ICD10) 可串自己的問卷

CY Pu*, YJ Chou (2016). The Impact of Continuity of Care on Emergency Room Use in a Healthcare System without Referral Management: An Instrumental Variable Approach. Annals of Epidemiology; 26(3):183-8 利用工具變項 探討在一個無轉診制度體系下，COC對ER的影響

Background Lower avoidable hospitalization Lower emergency room use (avoidable or unavoidable) More preventive care Lower medical costs Better adherence Higher satisfaction A few studies found good COC has no effect. Referral system(?): it is unreasonable to expect patients to visit the same physician for different diseases

Background Why COC may be endogenous? Personal traits Reverse outcome Some other unobserved confounders Objective: to test whether COC prevents emergency room use in a healthcare system without referral management, using an instrumental variable (IV) approach. Effects? Increase or decrease? Hypertension and type II diabetes.

Data and Methods IV: average continuity within a family of the same diseases. Similar health-care-seeking behavior, families share beliefs regarding health and behaviors related to illness. Family members’ health-care-seeking behavior do not directly affect whether a person visits the ER. Previous studies have used family member’s disease status as IV for self disease status. Close friend’s health behavior as IV for self health behavior. Ettner (1996), Xu (2002)  Regular doctor and preventive use. IV=length of residency.

Data and Methods The NHI claims data enabled us to determine whether a person has family members by determining whether the person has or is a dependent. Partial family member. A hypertension(diabetes) patient had a valid instrumented COC score if he or she had at least one family member with hypertension (diabetes) within the same year (averaged, excluding the person’s own COC score). 23 million NHI data (2008~2009). Excluded: <20 years, had <3 outpatient visits. Outcome variable: ER in 2009 (one of the ICD codes being hypertension/diabetes) Model: IV-Probit, COC treated as a continuous endogenous variable.

Tsai HY, YJ Chou, CY Pu* (2015) Continuity of Care Trajectories and Emergency Room Use among Patients with Diabetes. International Journal of Public Health;60(4):505-13. 糖尿病患COC對ER之軌跡分析

Background For all durations of the disease? A person may spend considerable time seeking the most suitable doctor before undertaking the long-term treatment of the disease. Studies found COC = patient satisfaction, and COC per se may be irrelevant to patient outcome (Guthrie et al, 2000) Two US studies: many patients considered another physician before choosing their current physician (Harris 2003, Tu et al 2008). The benefits of seeking the most suitable doctor by sacrificing care continuity are unknown.

Methods Data: 2005 NHI a million cohort. Newly diagnosed diabetes patient: ICD-9-CM 250.xx in 2005  2 outpatient + prescription in 2005, or 1 inpatient with main diagnose being 250.xx + prescription Outcome variable: the frequency of ER visits that occurred in the final 2 years of the 6-year follow-up period (2010 and 2011) Exclusion criteria: Dead prior to 2011, had less than 5 diabetes outpatient visits during 2006~2011. COCI: Moving average: (1-5)(2-6)(3-7)………. DM-outpatient: ICD-9-CM 250.xx +prescription

Model Trajectory analysis (trajectory assignment): to estimate a discrete mixture model for longitudinal data grouping by using censored normal distribution. Specifying different functional form model selection based on Bayesian Information Criteria (BIC) Model controlled for: Age, sex, insurable income, area, Charlson comorbidity, and diabetes severity (DCSI), level of institute for first visits. And then: Negative binominal models, outcome variable=number of ER visits

Results

Discussion Diabetic patients have different COC trajectories, and these trajectories have different patient outcome. Some plausible explanations: High maintainers have other unmeasured good health behavior. Those who found suitable physicians are less likely to switch. The first visit may or may not be pure luck. High COC per se has good effect, regardless of whether physician was suitable.

CS Hsu, YJ Chou, CY Pu* (2016). The Effect of Continuity of Care on Emergency Room Use for Diabetic Patients Varies by Disease Severity. Journal of Epidemiology, 26(8): 413–419. 照護持續性之效果是否受疾病嚴重度影響?

Background Does the positive effect of COC depends on the level of disease severity? Severity of disease and patients’ awareness of this severity predicts the patients’ treatment behavior (DiMatteo et al, 2007) For diabetic patients, treatment effects depend on severity and comorbidity (Gebhardt, 2013). After controlling for severity, there is no association between COC and improved monitoring for diabetic patients (Gill, 2003)

Methods Data: 2005 NHI a million cohort. Newly diagnosed diabetes patient: ICD-9-CM 250.xx in 2005  2 outpatient + prescription in 2005, or 1 inpatient with main diagnose being 250.xx + prescription Outcome variable: the frequency of ER visits each year (all ER and DM-specific ER) Diabetic severity: Diabetes Complications Severity Index (DCSI) A measure of the number and type of diabetes complications. These complications include retinopathy, nephropathy, neuropathy, cerebrovascular disease, cardiovascular disease, peripheral vascular disease, and metabolic disease (Young, et al 2008) .

Model Model: negative binominal model estimated using generalized estimation equations Included the “no-index” group (DM-outpatient<3) Several interactions: Age*COC Sex*COC DCSI*COC Charlson comorbidity*COC

Results

Discussion COC has harmful effect in addition to the severity effect when severity level reaches a certain level (DCSI ≥ 2 in this study) At a non-severe stage, patients can enjoy the positive outcomes because of less disruption. As the disease severity increases, the treatment or self-disease management may outweigh the importance of COC. However, COC is more beneficial for patients with higher comorbidity. Great knowledge on chronic disease management? The “No-index” group actually has best outcome. Actual adherence, first prescription was inappropriate.

GY Kuo, YJ Chou, CY Pu* (2017). Effect of continuity of care on drug-drug-interaction. Medical Care. 55(8):744-751 照護持續性與開錯藥(藥物交互作用)

Background Pharmacologic development has provided physicians with a growing number of strategies for countering various diseases. Before PharmaCloud was truly effective, do patients receive multiple drugs with interactions? Drug-drug interactions (DDIs) are a major category of adverse drug reactions. DDIs occur when a patient is prescribed >=2 drugs that have negative interaction effects. Elderly: comorbidity is common Objective: To evaluate whether patients with higher physician and site COC levels are less likely than those with lower levels to be prescribed drugs with known DDIs and whether this effect varies depending on comorbidity scores.

Methods DDI: the Phansalkar list which was formulated by a panel comprising experts on different aspects of the drug prescription system. Fifteen DDI types were considered according to the strict criteria established by these experts. Phansalkar S, Desai AA, Bell D, et al. High-priority drug-drug interactions for use in electronic health records. J Am Med Inform Assoc 2012,19:735-743.

Methods(con’t) CCI: The CCI categories are calculated on the basis of 17 disease categories: diabetes with diabetic complications, congestive heart failure, peripheral vascular disease, chronic pulmonary disease, mild and severe liver disease, myocardial infarction, solid tumor with or without metastasis, peptic ulcer disease, cerebrovascular disease, hemiplegia, renal disease, leukemia, lymphoma, metastatic tumors, and acquired immunodeficiency syndrome.

Methods(con’t) 4 observation intervals—1, 2, 4, and 9 years. People were included only if they remained alive throughout an entire observation interval. For example, if the observation interval was 1 year, only people surviving 1 year after the baseline were considered. Their COC and DDIs were then analyzed in that 1-year interval. The 2013 data are the most recent data available, representing events occurring 9 years from the baseline year.

Methods(con’t) Negative binomial regression. Incidence rate ratios (IRRs) were presented. DDIs were considered in the model as an actual count number. To account for the possibility that the level of COC is endogenous (ie, patients select themselves into different COC levels), we used the inverse probability of treatment weightings, which were estimated using a propensity score to balance the observed characteristics among different COC values. Age, sex, average number of medications prescribed, hospitalization frequency, outpatient site choice, CCI, and NHI enrollment category.

What’s next? Using NHI data can only tell so many stories…

Provider COC in pediatrics Provider continuity of care (COC) is closely related to patient outcome in pediatrics. Overwhelming studies have indicated that high COC leads to better patient outcome in pediatrics. Do parents know this? If they do, are they willing to make efforts to maintain good COC for their children?

Survey A cross-sectional survey was conducted between August 2017 and February 2018 across 4 hospitals in Taiwan.

Willingness to pay (money and time)

Self-reported reasons for changing pediatric physicians.

換醫生與跟對醫生? 持續地跟錯醫生或許更糟… Volume-outcome 已經有研究指出之間physician volume 及patient outcome之 因果關係 在眼科領域，有研究提出白內障手術量越多的眼科醫生 發 生術後不良事件的機率越低。 此外，也有學者探討volume-outcome relationship也出現在 疾病檢測，如Myocardial infarction(MI)或是乳癌的檢測。

研究目的 探討眼科醫生的服務量與照護持續性對於青光眼的檢測是否存在交互 作用。 **Specific Questions 好的照護持續性是否會有較高的檢測率? 高COC代表好的physician-patient relationship，且因經常回診而有追蹤 檢查 上述的COC效果(好的COC有好的patient outcome)是否會因醫師服務量 而有不同?

Study Design Nested Case-control Study Study period: 2007 - 2016 (Files: Health01全民健保處方及治療明細檔_門急診, Health04全民健保處方及治療醫令明細檔_門急診, Health07全民健保承保檔, Health81全民健保特約醫事機構資料檔) Data source: Taiwan National Health Insurance Research database. Confirmed glaucoma diagnosis ICD9/10CM 365.x, 364.22/H40.x, H42.x + Drug use at least 0.5-1 year (健保代碼) Study Population gender, age, socioeconomic status, continuity of care index, outpatient department visits, location, ownership, accreditation level (Kooner, AlBdoor, Cho, & Adams-Huet, 2008) Independent Variables

Flow Chart New case diagnosis of glaucoma With ICD code and drug use Study population N= 256,268 Approximately incidence rate: 11 ‰ NHIRD Whole population from 2007-2016 Exclusion: Angle-closure glaucoma 365.2x/H40.2x Uveitis 364.3/H20.9 Ocular trauma 365.65/H40.30x Vitreous hemorrhage 379.23/H43.13 Diabetic Retinopathy 362.01/E11.319 Retinal Vessel Occlusion 362.3x/H43.13 Any eyeball surgery 2007 new case(cannot trace back) Patients under 20 years old Subjects had missing data Early detection N= 256,147 Late detection N=121 Good Continuity of care Poor Continuity of care High Volume Low Volume Index date Continuity of care 1 year Poor outcome? Time from first confirmed diagnosis to use more than 3 or 4 medications or surgery or a poor outcome within one year. (Kooner, AlBdoor, Cho, & Adams-Huet, 2008)

Definition of Variables Outpatient visits Continuity of care index Ophthalmologist volume Interaction Frequency of ophthalmology OPD visit. Total number of visits 1 year prior to index date. Use continuity of care index to calculate and divide into 2 groups, good and poor COC. (visits that over 2 times) Analyze in volume of total patients, total glaucoma patients, and glaucoma patients as percentage of total patients. Interactions of above variables **Index date: The date that patient was detected as confirmed glaucoma.

Socioeconomic Status (NTD) Table 1. Characteristics of total observations Total Early detection (%) Late detection (%) N=256,268 N=256,147 N=121 Gender male 132,391(51.66) 132331 51.66 60 49.59 female 121,877(47.56) 121816 47.56 61 50.41 undefined 2,000(0.78) 2000 0.78 - Age AVG(SD) 56.11(16.46) 56.10(16.45) 63.4(17.21) 20-45 68358(26.67) 68342 26.68 16 13.22 46-65 109822(42.85) 109772 42.86 50 41.32 >65 78088(30.47) 78033 30.46 55 45.45 Socioeconomic Status (NTD) <=22800 143376(55.95) 143284 55.94 92 76.03 22801-45800 71237(27.8) 71215 27.8 22 18.18 >45801 41655(16.25) 41648 16.26 7 5.79 City North 130972(51.11) 130941 51.12 31 25.62 Central 53023(20.69) 52979 20.68 44 36.36 South 60107(23.45) 60072 23.45 35 28.93 East 12166(4.75) 12155 4.75 11 9.09

o Table 1. Characteristics of total observations (continued) Total Early detection (%) Late detection (%) N=256,268 N=256,147 N=121 Continuity of Care Index Not defined 137300(53.58) 137248 53.58 52 42.98 Low 35353(13.79) 35331 13.79 22 18.18 Medium 47847(18.67) 47817 18.67 30 24.79 High 35768(13.96) 35751 13.96 17 14.05 Volume Ophthalmology 84572(33.00) 84527 33 45 37.19 87117(33.99) 87067 33.99 50 41.32 84579(33.01) 84553 33.01 26 21.49 Glaucoma 84515(32.98) 84481 32.98 34 28.1 87217(34.03) 87167 34.03 84536(32.99) 84499 32.99 37 30.58 Percentage Distribution 84506(32.98) 25 20.66 87172(34.02) 87114 34.01 58 47.93 84590(33.01) 84552 38 31.4 Hospital Location North 128656(50.20) 128624 50.21 32 26.45 Central 53079(20.71) 53036 20.71 43 35.54 South 62216(24.28) 62181 24.28 35 28.93 East 12317(4.81) 12306 4.8 11 9.09 Ownership Public 39286(15.33) 39257 15.33 29 23.97 Private 216982(84.67) 216890 84.67 92 76.03 Accreditation Level Medical center 52333(20.42) 52312 20.42 21 17.36 Regional hospital 45368(17.7) 45321 17.69 47 38.84 District hospital 21299(8.31) 21287 8.31 12 9.92 Clinic 137268(53.56) 137227 53.57 41 33.88 o

結語 照護持續性的效果或許真的存在。 持續努力證明事實上不存在。 因果關係仍是重點。

Thank You