Saturday, August 22, 2020

Covid Test Interpretation and Response

I am seeing stories about people who are taking leave from work because they are infected with covid, and they are expected to return when they are well. So as a kind of ready-reference let me review what the decision process is for ending isolation, according to the CDC:
You can be around others after:
  • 10 days since symptoms first appeared and
  • 24 hours with no fever without the use of fever-reducing medications and
  • Other symptoms of COVID-19 are improving*

*Loss of taste and smell may persist for weeks or months after recovery and need not delay the end of isolation​
You will notice that there is nothing in that list about "getting a negative test result."

That is because negative test results don't really mean anything. They do not tell you whether you are sick or well.

One problem with developing an accurate test is determining the standard. To tell if the test works on somebody, you need to know if they are sick or not when you test them. How can you know that -- especially if they are asymptomatic? Researchers test and then test again the next day, to see how consistent a test is. They figure if you're sick the second day, you were probably sick the day before. They know they underestimate the number of errors this way but it is the best method they have.

A physician interpreting the test takes prior probability into account. For instance, they will consider the infection rate in the region. They will adjust their prior probability estimate if you have traveled to a place with a high rate, or have engaged in risky behaviors like going out without a mask or attending a concert or party, have been exposed to a sick person, or if you have symptoms of covid-19. Then they take the test result and make a diagnosis. The Bayesian math is more than we need here, but the point is that they do not just look at the test result.

The positive result is pretty good on its own -- if the test sees enough viral RNA to trigger a reaction then the result is positive. It seems that positive test results are about 98 percent accurate; if it says you're sick, you're sick.

But what if the result is negative? A false negative test has serious consequences, for instance it can lead an infected person to interact with others with minimal precautions. Because there is no true gold standard, researchers can only estimate the false negative rate. One review of five studies estimated false negative rates up to 29 per cent, but those estimates have been questioned and are probably low.

The point is that a negative test result does not mean you are not infected.

The BMJ (formerly British Medical Journal) has published guidelines in an article titled, sensibly, "Interpreting a covid-19 test result." Their interest is professional, they are mostly advising doctors about when they can return to their practice without infecting their patients. Here's what they say:
While positive tests for covid-19 are clinically useful, negative tests need to be interpreted with caution, taking into account the pre-test probability of disease. This has important implications for clinicians interpreting tests and policymakers designing diagnostic algorithms for covid-19... False negatives carry substantial risks; patients may be moved into non-covid-19 wards leading to spread of hospital acquired covid-19 infection, carers could spread infection to vulnerable dependents, and healthcare workers risk spreading covid-19 to multiple vulnerable individuals. Clear evidence-based guidelines on repeat testing are needed, to reduce the risk of false negatives.

The worrisome thing to me is that employers are using the test as a binary decision-maker. If your test is positive, then they let you stay home, but if it's negative they interpret that to mean that you should get back to the office and get to work. That kind of decision is going to send a lot of sick people back into a risky environment, which will in turn drive up the numbers, put more people in danger, and perpetuate the chaos of this pandemic.

In the face of a negative test result, a responsible physician should make an informed diagnosis.

Here is how the BMJ piece suggests doctors explain it to patients:
  • No test is 100% accurate
  • If your swab test comes back positive for covid-19 then we can be very confident that you do have covid-19
  • However, people with covid-19 can be missed by these swab tests. If you have strong symptoms of covid-19, it is safest to self-isolate, even if the swab test does not show covid-19

We should play this safe. Supplement a negative test result with a medical diagnosis. If you are sick, remain isolated until you meet the CDC criteria given above.

Monday, August 10, 2020

MoCo Popo Changes Being Discussed

If you do the math, you see that approximately half the adult men in the United States have a criminal record, meaning they have at least been arrested. Generally those are not the ones who are running things, they are not politicians, business executives, lobbyists. Our society is two-tiered, where one tier locks the other tier up. If you're on the locking-up side of it, you may be unaware of this.

It is not news to note that the Black community is especially targeted by law enforcement. In 2010, one-third of Black men were reported to have a felony conviction, and the percentage was growing. Black Lives Matter protests across the country are largely about police bias and violence.

Montgomery County is mostly known as a wholesome suburban community, diverse and affluent. Occasionally a police problem reaches the public eye, but generally we are a peaceful, liberal place. The county recently released a Report on Local Policing Data and Best Practices, and the results reveal some issues.

I should note that Montgomery County, Maryland, is probably more self-aware and less racist than many places. We are an extremely diverse county, progressive, welcoming. So as you read results from this report, remember that a whole lot of the country is a whole lot worse. Montgomery County has begun studying the extent of their problem, at least, and is discussing solutions where problems are found.

The report is really about data, especially data that focuses on "constitutional" policing practices (bias-free policing and use of force) and police-community relations (police satisfaction, trust in police, and police legitimacy). It contains a number of tables and a lot of discussion of datasets and links to them. I did not open any and analyze but it could be a fun quarantine activity. We are not necessarily interested here in databases so much as what the data can tell us about our county.

Most of the results in the report have been broken down to highlight differential treatment of ethnic groups, broadly defined. TL;DR : the cops are harder on Black people than anybody else.

The county has data on traffic stops but not on pedestrian stops or street stops that are not done in response to 911 calls. That is, stop-and-frisk and ordinary hassling remain off the record. For traffic stops, 2018 data shows (percents in parens calculated by me):
  • Black people accounted for 18 percent of all residents v. 32 percent of MCPD traffic stops (177% of equity rate)
  • White people accounted for 44 percent of all residents v. 35 percent of MCPD traffic stops (79.5% of equity rate)
  • Latino people accounted for 19 percent of all residents v. 20 percent of MCPD traffic stops (100.5%, about what is expected)
  • Asian people accounted for 15 percent of all residents v. 7 percent of MCPD traffic stops (46.7% of equity rate)
County records have no data to determine whether there are disparities in MCPD searches of pedestrians, but searches at traffic stops were conducted on 4.4 percent of Black drivers, compared to 3.3 percent of Latino drivers, 2.0 percent of White drivers, and 1.3 percent of Asian drivers. This is what happens when cops use their personal judgment to decide if someone looks sketchy -- Black drivers get searched more than twice as much as white ones.

Why would you search somebody at a traffic stop in the first place -- to see if they're hiding their registration sticker in their pocket? If they are breaking a traffic law, then ticket them for that. If there is contraband in plain sight, well that's just dumb and no search is necessary. Otherwise, all of these searches mean that a traffic cop "suspected" something.

The county is planning to automate more traffic enforcement, so if you're speeding with a bag of weed in the glove compartment you get a ticket for speeding. I hate traffic cameras myself but you have to admit they are blind to privilege.

"Hassling" should be a legal term.

Traffic stops resulted in arrests of 2.2 – 2.3 percent of Latino and Black drivers compared to 1.3 percent of White drivers and less than one percent (0.9%) percent of Asian drivers. These are people who are pulled over for a traffic offense and end up going to jail. Unless there is an active warrant for the driver, this should almost never happen. (BTW, note that arrests correlate with searches. This is a time when it is actually true that "if you do more testing you're going to find more cases.")

BTW if you are thinking, well those white people just weren't doing anything wrong, I can't help you.

How about the use of force?
  • Black people accounted for 18 percent of all residents v. 55 percent of use of force incidents (305% of equity)
  • White people accounted for 44 percent of all residents v. 26 percent of use of force incidents (59%)
  • Latinx people accounted for 19 percent of all residents v. 18 percent of use of force incidents (94.7%)
  • Asian people accounted for 15 percent of all residents v. 1 percent of use of force incidents (6.7%)
These numbers eerily confirm common stereotypes: Black percentages more than twice as high as white; force used on Blacks in Montgomery County more than three times their proportion in the population. BTW- the word "violence" appears here only once, in the term "Domestic disturbance/violence." It is never associated with police behavior in the report.

Looking at traffic violations, Black drivers were ticketed in 2019 at the rate of 321 per 1,000 population -- nearly a third of Black drivers got pulled over and ticketed in one year. The rate for white people was less than half that, at 132 per 1,000. Latinos were in-between, at 215/1,000, and Asians at the bottom of the list, a common thread here, at 70 violations per 1,000 population. Oddly, the "Other" group got a lot of tickets, at 319 per 1,000. I don't know what that is.

Another table in the report shows that Latino and Black drivers were more likely to earn six or more violations during a single traffic stop than any other racial and ethnic group. Are you seeing a trend here? Black and Latino drivers, especially males, are subjected to searches at higher rates than other groups. In 2019, 27 percent of Black adults in the county experienced a traffic stop compared to 17 percent of Latino adults, 14 percent of White adults, and 7 percent of Asian adults -- Blacks nearly twice as likely as whites. Further, 4.4 percent of Black drivers were searched in 2018 compared to 3.3 percent of Latino drivers, 2.0 percent of White drivers, and 1.3 percent of Asian drivers -- Blacks more than twice as likely as whites. Black men were three times as likely to receive any violation, to receive a citation, or to receive a repair order as white men, and twice as likely to receive a warning. This is in diverse, liberal Montgomery County.

The report details and links to a number of databases but acknowledges frequently that particular variables are not collected, or that data from various databases has not been linked. So for instance if someone is in a traffic accident, they do not record the race or ethnic group of the driver. It's on their license, but apparently they cannot connect the two.

Most of the recommendations in this report had to do with getting better data. We can tell enough from this to start working on an obvious problem, but the county needs better metrics and better ability to analyze them. The Washington Post says:
County council member Hans Riemer (D-At Large) has commissioned a study to figure out whether – and how — Montgomery might be able to move certain traffic enforcement functions out of the police department and into other government agencies... 
County Executive Marc Elrich (D) said he is “more than concerned” about the racial inequities in traffic enforcement and will consider the possibility of reallocating resources. 
Montgomery appears to be the only jurisdiction in the Washington metro region targeting traffic enforcement in policing, though the District last year moved automated enforcement from police to the transportation department... 
Officials say the differences signal discriminatory policing, though the police union and Police Chief Marcus Jones dispute this. 
“There are many factors that need to be considered beyond simple census data to determine if racial bias exists in law enforcement,” Jones said in a statement. Should police be in charge of traffic enforcement? In a suburb beset by racial inequities, lawmakers aren’t sure.
The US has been struggling lately with the concept of democracy, of providing rights and security to all its citizens, and bias in policing is central to the problem. Many Americans leave their house not knowing if they will make it back again, because of the police. Philandro Castile had been stopped by police forty-six times before they finally killed him. As NPR noted, "only six of them were things a police officer would notice from outside a car — things like speeding or having a broken muffler."

Again, if you think he was getting stopped for doing something wrong -- I'll give you six of those. What about the other forty times he was stopped? As this county report shows, Black citizens are targeted by police compared to white people, even here in progressive, prosperous Montgomery County.