Page 7 - SFHN1218pagesFINALSP.qxp_Page layout

Year-end Payroll and Accounts Payable

Considerations in Healthcare

With year-end 2018 fast approach- income for the years 2018 - 2025. 1099 are due by February 28, 2019 for
ing outlined below are recent 2019 There are certain exceptions for mili- paper filers and April 1, 2019 for elec-
IRS inflationary adjustments and also tary personnel and reimbursements tronic filers (required if more than
changes as a result of the Tax Cuts for expenses incurred in 2017. 250 Forms). A 30 day extension may
and Jobs Act (“Act”) that affect our • Certain employee achievement be requested via Form 8809. Beware
compliance with payroll and accounts awards can be excluded from employ- that payments for medical services
payable reporting. ees’ wages; however, the Act excludes always require a Form 1099 no matter
The Internal Revenue Service certain tangible property such as cash, the type of payee.
(“IRS”) recently released cost of living gift cards tickets to shows and sport- In addition the backup withholding
adjustments that affect retirement ing events. tax rate was reduced from 28% to 24%
related limitations for 2019. Some of • Certain meals and entertainment starting in 2018. Back up withholding
those changes include the following: expenses are disallowed; however, the can apply to payments such as divi-
• Contribution limits for employees 50% deduction for business related dends, interest, rents, commissions or
who participate in certain retirement BY LINDA GNESIN, CPA meals was retained. fees for services. Back up withholding
plans such as IRC §401(k), §403(b) Tax-exempt organizations should may be required if the following con-
and most §457 plans increased from also be aware of the new 21% excise ditions exist, including but not limit-
$18,500 in 2018 to $19,000 in 2019; to $56,000 in 2019. tax on excess executive compensation ed to, failure to provide a taxpayer
the catch up contributions for those The maximum earning subject to paid to covered employees in excess identification number (“TIN”), pro-
age 50 and over remains at $6,000. social security tax were increased of $1 million. In addition, there is a viding incorrect TIN, underreported
• The limit on IRA contributions from $128,400 to $132,900 for 2019. potential requirement to include cer- interest or dividends or failure to cer-
increased from $5,500 to $6,000 in Rates for Social Security and Medicare tain employee fringe benefits, such as tify that they are not subject to back-
2019; the catch up contribution for have remained the same; however, qualified transportation and parking up withholding.
those age 50 and over remains at included in the Act are personal as unrelated business income subject
$1,000. income tax rate changes for individu- to a 21% corporate tax rate. For more information contact Linda
• The limitation on the annual ben- als, which affected the amount of 2018 Forms 1099-MISC are gener- Gnesin, CPA, Manager,
efit for a defined benefit plan employee withholding in 2018. ally due to recipients by January 31, WithumSmith+Brown, PC,
increased from $220,000 to $225,000 In addition, the Act included vari- 2019. If there is reporting of nonem- at (973) 532-8867 or
in 2019. ous changes to fringe benefits, such as ployee compensation in Box 7, filing lgnesin@withum.com or visit
• The limitation for defined contri- the following: is also due by January 31, 2019. For www.withum.com.
bution plans increased from $55,000 • Qualified moving expenses can- other than Box 7 reporting, Forms
not be excluded from an employees’

Researchers Detect Medicare Deception

by Programing Computers to Detect Fraud

BY LISA BIANCO improve fraud detection. specialty and determined whether the cases. The researchers found the “sweet
Providers could lighten the predicted specialty differed from the spot” for detecting Medicare fraud to be
Twenty percent of workload for auditors and actual specialty, as indicated in the a 90:10 distribution of normal vs. fraud-
health care spending in investigators. Medicare Part B data. ulent data.
the United States comes Researchers from Florida Taghi M. Khoshgoftaar, Ph.D., co- “Our goal is to enable machine learn-
from Medicare, the pri- Atlantic University’s (FAU’s) author and Motorola Professor in FAU’s ers to cull through all of this data and
mary health care cover- Department of Computer Department of Computer and Electrical flag anything suspicious. Then, we can
age for Americans 65 and Electrical Engineering Engineering and Computer Science alert investigators and auditors who will
and older. Rumors and Computer Science exam- explains, “For example, if a dermatolo- only have to focus on 50 cases instead of
abound as to how much ined the Medicare Part B gist is accurately classified as a cardiolo- 500 cases or more,” says Bauder.
fraud exists in Medicare. dataset from 2012 to 2015. gist, then this could indicate that partic- Khoshgoftaar explains, “The goal is to
Authorities estimate that They focused on detecting ular physician is acting in a fraudulent or build a predictive model and create a bet-
yearly about $19 billion fraudulent provider claims wasteful way.” ter methodology for federal auditors. The
to $65 billion is lost to Dr. Taghi M. Khoshgoftaar within the dataset of 37 mil- For the study, Khoshgoftaar, senior fact that we are the first to use big data to
Medicare fraud, waste lion cases. Cases labeled as author Richard A. Bauder, Ph.D., student uncover Medicare fraud is very impor-
or abuse. “fraud” include patient abuse or neglect and data scientist at FPL and their Ph.D. tant. If you are lucky enough to be the
Human auditors or investigators and billing for services not rendered. student collaborators had to address the first, other researchers come to you. So
painstakingly check thousands of Physicians and other providers who com- high imbalance of the original labeled big far we have touched only the surface of
Medicare claims manually for specific mit fraud cannot participate in federal dataset. This occurred because non- this problem, really just uncovered the
patterns that may indicate foul play or health care programs like Medicare. fraudulent providers far outnumber tip of the iceberg.”
fraudulent behaviors. The U.S. The researchers aggregated the 37 mil- fraudulent providers. This is problematic This detection method also has appli-
Department of Justice reports that today’s lion cases down to a smaller dataset of 3.7 for machine learning approaches. The cations for other types of fraud including
fraud enforcement efforts depend mostly million. They devised a unique process to algorithms attempt to distinguish insurance, banking and finance. The
on health care professionals revealing map fraud labels with known fraudulent between the classes, but one dominates researchers are currently adding other
information about Medicare fraud. providers. Medicare Part B data includes the other and fools the learner. Medicare-related data sources such as
The journal Health Information provider information, average payments The researchers solved this problem by Medicare Part D.
Science and Systems recently published a and charges, procedure codes, number of using random undersampling, reducing Dean of FAU’s College of Engineering
study which is the first to employ procedures and the medical specialty, the dataset from 3.7 million cases down and Computer Science, Stella Batalama,
advanced data analytics and machine known as the provider type. to about 14,000 cases (for the best detec- Ph.D., foresees further impacts this
learning with big data from Medicare To obtain exact matches, the tion results). They created seven class research may have. “The methodology
Part B. The study’s aim was automating researchers used the National Provider distributions and used six different learn- being developed and tested in our college
the fraud detection process. Identifier (NPI) — NPIs are issued by the ers across class distributions from severe- could be a game changer for how we
Machine learning is a branch of artifi- federal government to health care ly imbalanced to balanced. The learning detect Medicare fraud and other fraud in
cial intelligence based on the idea that providers — to match fraud labels to the algorithm RF100 (Random Forest) was the United States as well as abroad.”
computer systems can learn from data Medicare Part B data. Researchers direct- the best at detecting the positives of
and identify patterns. In this study com- ly matched the NPI across the Medicare potential fraud events. Interestingly, For more information contact
puters were programmed to predict, clas- Part B data, flagging any provider in the keeping more of the non-fraud cases Gisele Galoustian at ggaloust@fau.edu,
sify and flag potential fraudulent events. “excluded” database as being “fraudu- helped the learner/model better distin- Media Relations Director
This approach could significantly lent.” They classified a physician’s NPI or guish between the fraud and non-fraud at Florida Atlantic University.

South Florida Hospital News southfloridahospitalnews.com December 2018 7