Author: akshay

 Amar Jesani shared in mfc group link to an article about v-safe data release.

The actual data could be downloaded from this website called icandecide.

The 5GB file can be extracted with p7zip to a 25GB CSV file.

$ md5sum consolidated_health_checkin.zip
53ff7a8153f44eaab4166f722b726fe1  consolidated_health_checkin.zip
$ md5sum consolidated_health_checkin.csv
345cf6ca148832141260aab8638bf0dc  consolidated_health_checkin.csv

$ wc -l consolidated_health_checkin.csv
144856044 consolidated_health_checkin.csv

(That’s 144 million records in this CSV file)

$ head -n 5 consolidated_health_checkin.csv
SURVEY_STATIC_ID,REGISTRANT_CODE,RESPONSE_ID,STARTED_ON,STARTED_ON_TIME,DAYS_SINCE,ABDOMINAL_PAIN,CHILLS,DIARRHEA,FATIGUE,FEELING_TODAY,FEVER,HAD_SYMPTOMS,HEADACHE,HEALTH_IMPACT,HEALTH_NOW,HEALTH_NOW_COMPARISON,VACCINE_CAUSED_HEALTH_ISSUES,HEALTHCARE_VISITS,ITCHING,JOINT_PAINS,MUSCLE_OR_BODY_ACHES,NAUSEA,PAIN,PREGNANT,PREGNANCY_TEST,RASH_OUTSIDE_INJECTION_SITE,REDNESS,SITE_REACTION,SWELLING,SYSTEMIC_REACTION,TEMPERATURE_CELSIUS,TEMPERATURE_FAHRENHEIT,TEMPERATURE_READING,TESTED_POSITIVE,TESTED_POSITIVE_DATE,VOMITING,DURATION_MINS,PREFERRED_LANGUAGE
vsafe-0-day-dose1,222-10271-84782,s244305050865137831057660547899056617007,12/31/2020,4:55:13 PM,0,,,,,Good,No,,Mild,N/A,,,,,,,,,Mild,,,,,Pain,,Headache,,,,,,,,English
vsafe-0-day-dose1,222-10325-02776,s258811629454233188277362395339553379505,05/19/2021,3:16:15 PM,0,,,,,Good,No,,,N/A,,,,,,,,,,,,,,None,,None,,,,,,,0.85,English
vsafe-0-day-dose1,222-10368-05218,s256518678527351061889187968276580937945,04/27/2021,4:11:31 PM,0,,,,,Good,No,,,N/A,,,,,,,,,,No,,,,None,,None,,,,,,,0.72,English
vsafe-0-day-dose1,222-10453-23273,s245552707728162053684731534374544736656,01/12/2021,3:31:16 PM,0,,,,,Good,No,,,N/A,,,,,,,,,,No,,,,None,,None,,,,,,,,English

As you can see there are many columns, which we will have to decode.

Combing through the whole file again and again is taking a lot of time on my computer. So I decided to write a python script that’ll do all analysis in one pass of the file.

But that was taking even more time.

So I decided to put this data into postgreSQL to do the analysis.

$ sudo -u postgres createuser health

$ sudo -u postgres createdb vsafe -O health

$ cat load.sql
SET datestyle TO dmy;
CREATE table if not exists checkin (
        SURVEY_STATIC_ID varchar, — eg: vsafe-0-day-dose1
        REGISTRANT_CODE varchar, — eg: 222-10271-84782
        RESPONSE_ID varchar,
        STARTED_ON DATE,
        STARTED_ON_TIME varchar,
        DAYS_SINCE int,
        ABDOMINAL_PAIN varchar,
        CHILLS varchar,
        DIARRHEA varchar,
        FATIGUE varchar,
        FEELING_TODAY varchar,
        FEVER varchar,
        HAD_SYMPTOMS varchar,
        HEADACHE varchar,
        HEALTH_IMPACT varchar,
        HEALTH_NOW varchar,
        HEALTH_NOW_COMPARISON varchar,
        VACCINE_CAUSED_HEALTH_ISSUES varchar,
        HEALTHCARE_VISITS varchar,
        ITCHING varchar,
        JOINT_PAINS varchar,
        MUSCLE_OR_BODY_ACHES varchar,
        NAUSEA varchar,
        PAIN varchar,
        PREGNANT varchar,
        PREGNANCY_TEST varchar,
        RASH_OUTSIDE_INJECTION_SITE varchar,
        REDNESS varchar,
        SITE_REACTION varchar,
        SWELLING varchar,
        SYSTEMIC_REACTION varchar,
        TEMPERATURE_CELSIUS varchar,
        TEMPERATURE_FAHRENHEIT varchar,
        TEMPERATURE_READING  varchar,
        TESTED_POSITIVE varchar,
        TESTED_POSITIVE_DATE varchar,
        VOMITING varchar,
        DURATION_MINS FLOAT,
        PREFERRED_LANGUAGE varchar
);

COPY checkin FROM ‘consolidated_health_checkin.csv’ DELIMITER ‘,’ CSV HEADER

$ psql -U health vsafe -f load.sql
…
COPY 144856043
 

That took about 25G space as well.

Beautiful. Now we can do all kinds of querying.

Actually, not yet. There’s one more thing we have to do. Create some indexes for making queries easier.

CREATE INDEX checkin_health_impact_idx ON public.checkin USING btree (health_impact);

Now, there are some issues with this data. For example:

ERROR: could not create unique index “checkin_pk”
  Detail: Key (response_id)=(s252082802016465320050574992159464366472) is duplicated.
  Where: parallel worker

response_id is duplicated, although it looks like every response might be unique.

But let’s ignore that now for an interesting query result:

 

That’s the distribution of the Health Impact column. 81 million responses say N/A, 56 million responses include no value (null) for this column  and the tail kind of begins there.

When I do select count(distinct(registrant_code)) from checkin; I get 9,552,127 which means only 9.5 million registrant_codes are included in the dataset. Since v-safe allows adults to respond on behalf of children, it is probably likely that there are more individuals in the dataset than the registrant_codes.

Then I ran select count(distinct(registrant_code)) from checkin where health_impact like ‘%Get care from a doctor or other healthcare professional%’; and it returned 797,396. Which means at least 797K people checked this option (with or without other options)

Now let us look at the variable of interest, healthcare_visits. The query I ran is select healthcare_visits, count(*) from checkin where health_impact like ‘%Get care from a doctor or other healthcare professional%’ group by healthcare_visits ;

The result is

 

Note that I haven’t deduplicated by registrant_code here. 

So I tried a different query: select count(*) from checkin where healthcare_visits  like ‘%Hospitalization%’; the answer to which is 83,690.

Let us try deduplicating by registrant_code on that:

select count(distinct(registrant_code)) from checkin where healthcare_visits  like ‘%Hospitalization%’; returns: 71,911

Which means, there’s some amount of duplication in the row data as to registrant_codes and reports. In other words, from the same registrant_code, you can have multiple reports of Hospitalization.

This data is rather messy and I’m not exactly sure how icandecide is arriving at “individual” in their numbers because all I see are registrant_code.

Now, on to some more interesting stuff. What is the distribution of systemic_reaction in registrant_codes who reported Hospitalization?

select systemic_reaction, count(distinct(registrant_code))from checkin where healthcare_visits  like ‘%Hospitalization%’ group by systemic_reaction ;

That turned out disappointing because the result was 68,170 NULL fields.

 

 But among the non-null fields, “None”, “Fatigue or tiredness”, “Headache”, etc are leading. (Do note that this is a multi-value column and there could theoretically be a symptom that appears in the tail of this column multiple times thus occurring more number of times than these ones.)

I also looked at the other files available for download.

It seems like the Consolidated_health_checkin_u3[1].zip must be under 3 children. The consolidated_registrants[1].zip file makes me think that each registrant_code actually uniquely identifies an individual. Because children are having separate registrant_code with guardian registrant_code mapped in this file. The other files are about race/ethnicity and vaccine that was administered.

 

The under 3 file includes 116394 reports. Some of the discrepancies in number between my analysis and ICAN’s dashboard probably comes from them adding both these together. 

If you look closely, many theories of change in public health where technology is involved has, at its heart, the following idea:

Adopting Technology -> leads to -> Better Health

This is a meaningless assumption guided by the hype around what technology can accomplish and the wishful thinking on solving large problems.

Firstly, technology is a large, amorphous, heterogenous categorization of human innovations. There are thousand different kinds of technology. One could say anything that human beings have made is technology.

Wikipedia says: “Technology is the sum of any techniques, skills, methods, and processes used in the production of goods or services or in the accomplishment of objectives”

Oxford dictionary says: “scientific knowledge used in practical ways in industry, for example in designing new machines”

Here is a list of things. Pick the ones you think are technology:

• MRI machine
• Stethoscope
• Instant messaging app like WhatsApp
• A paper tea cup
• Polio vaccine
• Mobile phone
• Solar panel
• Wallet
• Pen
• Clothes
• Fishes
• Scissors
• Fan
• Ubuntu linux
• Paracetamol
• Breast milk

Generalizations like “adopting technology will improve universal health coverage” are as useful as saying “human innovations will improve universal health coverage”.

The second problem with the unquestioning acceptance of technology is that technology isn’t always positive, or even value neutral.

• Nuclear bomb
• Pegasus spyware
• Deepfake
• Fake news bots
• Addictive apps
• Fossil fuel
• Heroin

Now if you are a tech-bhakt, your primary reaction to the above list is “Oh, but you know, these are just harmful uses of an otherwise good/neutral technology. It is a human problem, not the technology’s problem.” But please read on carefully.

The world we live in is populated by an increasing number of human beings. And human beings interact with technology in many ways. Some are predictable, some are unpredictable. The effect that a technology has on anything cannot be assumed to be “universally positive”. That effect has to be studied and understood.

That is not an argument against developing technology. It is an argument only against how technology is advertised and incorporated into human life. Technology should not be pushed into systems without weighing the potential advantages and potential harmful effects it can have. Such push can be counter productive if the real harms outweigh the real benefits.

Any use of technology will lie in a spectrum that ranges from extremely beneficial to extremely harmful. It takes discretion to identify where on the spectrum it lies. That human discretion, rationality, and scientific temper is what we need to develop in theories of change surrounding technology in public health.

   ***

Now that we have accomplished that technology in public health needs to be evaluated on an intervention-by-intervention basis, we can look at some specific examples.

Digital (enabled) delivery of healthcare

This vague concept is a slice of the “technology” concept we discussed above. Digitally enabled healthcare delivery can mean anything. Is it digitally enabled if I use a digital thermometer or a digital blood pressure device? Is it digital delivery if I’m doing a consultation over WhatsApp video call?

Let’s take a plausible example. A hospital information management system with electronic medical records and teleconsultation. This probably is something that many people have in mind when they’re talking about things like “medical and diagnostic connectivity throughout life course for every individual” or “sophisticated early warning systems leading to better preventative care”.

That example brings up a lot of questions. Which hospital are we talking about? Where is it located? Who are the beneficiaries and users of these? What kind of skills are we talking about? What kind of resources are available in these settings? What costs in terms of attention, time, effort, fatigue, etc are involved in utilizing these systems? What kind of software is available? How practical are the benefits? What are the challenges in taking data out of an EMR system and building early warning systems out of it? What foundational technologies are we lacking to build such systems? How will data from EMRs be analyzed? Who will do the analysis? What are the political processes occurring in India which could be connected to these? What are the incentives given to private sector in this? What are the protections required for patients? What are the support structures required for healthcare workers? Who is this intervention aimed to benefit? How does it affect health equity? Is it solving a problem that has been expressed by the community that it is being incorporated in?

These and countless other questions have to be answered before considering whether an intervention like above will lead to the impact that it is assumed to produce. Now, as is evident from these questions, the answers will vary widely depending on the settings. It might (or might not) produce an overwhelmingly positive impact in a super-specialty hospital in Koramangala for a software developer working with Infosys. It might not produce a similar impact in a PHC in Koppal for a NREGA dependent person. Unfortunately a lot of Indians are of the latter kind.

Techno-legal regulations

Here is another vague slice. Of course technology has to be regulated. Technology has always been regulated. It is just some newer technologies which are only slowly getting regulated. Things like databases and software platforms. The concern that regulations try to address in here are “Citizen rights, privacy and dignity”, “reducing technological inequities, algorithmic bias”, etc.

But “regulations”, like “technology” are not a sure-shot solution to anything. A lot of regulations stifle technology but doesn’t help fulfill the purpose it was meant to be either. Take the telemedicine guidelines released in March 2020, for example. In an attempt to enable telemedicine, it restricted the kind of diagnoses and prescriptions that can be made over telemedicine.

Getting regulations right is super hard. In the case of software based technology, even when regulations are right and tight, people tend to find loopholes rather quickly. Because software can quickly be adapted, it is possible to follow regulation and still continue doing bad stuff. Take how after GDPR came into place requiring consent for cookie use, so did dark patterns in cookie consent pop-ups.

India has a government which went to the Supreme Court to argue that privacy is not a fundamental right. When the government itself is involved in treating human beings as citizens to be controlled through surveillance, what insulation can regulations provide to human rights like privacy?

In other words, “equitable, people-centered and quality health services” and “improved accountability and transparency in the health system” cannot come through techno-legal solutions when the democracy does not have those in its priority list. Surely, technology and law can be instruments of social transformation. But only in the right hands.

There is no question of equity and people-centeredness emerging out of a process that does not have representation of people in it. What about quality? There are already some frameworks for quality in healthcare service. NHSRC, NABH, etc have various accreditation policies for hospitals. It takes a lot of work to build a culture of quality in a complex organism like a hospital, let alone health system. Culture is not something technology can fix.

Technology can be omnipresent. But a human cannot yell at a machine to get justice. That technology can lead to better accountability is a dream. The game where technology rapidly adapts to regulations and finds loopholes – human beings are 10 times better than machines in that game. Any accountability system based on technology will be gamed by human beings.

To see how technology and law affects transparency, one just has to look at what is happening to Right To Information act in our country today. No matter how “sophisticated” our technology gets, human beings are going to remain human beings.

   ***

And that is where “trust in the health system” comes in. How should human beings trust a system that doesn’t listen to them, negates their experiences, puts barriers in front of them in accessing healthcare, reduces health to the singular dimension of curative services (or recently vaccinations), treats them as undeserving, and regularly intrudes and violates their right to life and bodily integrity? What app should they install to download some trust?

Discussions on technology in public health need to wait till we discuss who our health systems are for. And once we have an answer on that, we should invite those people to the table. And when they state the problems that they face in leading a healthy life, those are the problems to be solved. Work backwards from there and you’ll realize that a lot of what we have are problems that don’t require technology to command citizens, but instead require human beings to listen to human beings.

At least 300 million people in India live below poverty line. And that line is drawn somewhere around an income of ₹1000-1500 per month. If we draw the line double that, the number of poor also doubles.

That’s the bottom of the bottomless pyramid.

Half a billion people who earn less than ₹3000 a month.

If you earned that much, what would your priorities be? Food? Shelter? Financial security? Education for a child?

What about your own health? 

Imagine you have diabetes too. The cheapest food you have all around you is rice or wheat based. If you want to decrease carbohydrates and not go hungry, how much can you spend on food? And if your sugars are not under control, would you spend more on a combination of multiple oral hypoglycemic agents that might cost about ₹500 per month?

   ***

Scraping the bottom of the pyramid works beautifully in consumer goods. You build something dirt cheap for the poor. Take a ₹2 shampoo sachet. You can cut down the size of the sachet to make it even cheaper.

You can’t sell half a metformin tablet to a poor diabetic.

You can’t prescribe a 1 day course of antibiotic.

You can’t cure pain with an injection.

But you can. Indeed that’s the kind of healthcare that those at the bottom of the pyramid currently receive. Sub-standard, inappropriate, and incomplete.

Because healthcare, unlike consumer goods, doesn’t become cheaper at the bottom of the pyramid. It actually becomes more expensive due to the intersection of vulnerabilities.

   ***

There is simply nothing to scrape at the bottom of the pyramid for healthcare.

Someone else has to pay.

A third party.

Could be the government. Could be charity. Could be grants.

But hey! If someone is paying, does it matter whether it is the beneficiary or a third party? 

   ***

That’s the logic with which most NGOs in health and government facilities work.

Say you’re a doctor in a PHC. The government pays you. You deliver healthcare to the poor. Simple economics.

Where does the government get money? It raises money through taxes, etc.

What if you’re a non-governmental organization? You get donations/grants in what is called “fund-raising”.

(There’s of course a cross-subsidization model which may look different superficially, but isn’t very different in the larger scheme of things)

Is this any different from first party payment?

Very different!

   ***

The first issue is that of accountability. Accountability lies where money flows from. If my healthcare is paid for by someone else, my healthcare provider isn’t accountable to me.

Public health facilities are not accountable to the poor that it serves healthcare to. They are only accountable to the hierarchy above them.

NGOs are not accountable to the poor that they serve healthcare to either. They are only accountable to funders. (Typically NGOs which are able to diversify their funding source is able to decrease the power that funders have to some extent by dividing the funders into many).

Why, though? Because accountability without control doesn’t work.

If you want to hold someone accountable, you have to be able to control them in some way.

   ***

When there is no accountability, the next issue is that of quality.

In first party payment, quality assessment is decentralized. Every individual makes their own assessment about the quality of care they receive. And this instantly translates to payment, recurring visits, etc.

In third party payment, quality assessment is different. It uses “metrics”. And metrics are difficult. Funders typically look at fancy metrics like “decrease in maternal mortality rate”. The problem with such “key” metrics are that they capture very little nuance and sometimes no meaning.

To government, for example, where the whole hierarchy is just supplying metrics to someone else, it becomes a complete number game. (Recommended reading: Chasing Numbers, Betraying People)

To NGO funders who have a bit more involved staffing structure it goes beyond numbers to also include “reports” filled with presentation-worthy photographs.

It no longer matters whether the individual receives quality healthcare as long as the metrics and reports are looking good.

   ***

Now let us look at something totally different. The CSR sector spent about 2600 crore rupees in health in 2020-21 FY. That’s about 1% of India’s national health budget. As per national health accounts 2017-18, the combined contribution of NGOs, corporates, foreign aid, etc to India’s health expenditures is less than 10%. 

By all means, the government is the single largest provider and payer of healthcare for the bottom half of India’s pyramid.

   ***

If you read all of this together, there are certain insights to be gained about why certain things are the way they are.

Why do NGOs build/research “models”? Because the kind of money it takes to deliver care to a population larger than what “model”s serve is hard for NGOs to come by.

Why does everyone want to build software? Because software can (theoretically) “scale” to large populations without a lot of money.

Why do NGOs focus on showcasing “reach”? Because numbers mean impact for funders. And creating the impression of quality is more important than quality.

Why does public health system get away with delivering poor quality healthcare? Because there’s no real way citizens can hold health system accountable. The constitutionally mandated way they can do so has been hijacked by issues like religion, party, and war.

   ***

What to do about all this?

1. Look deeper than numbers – everywhere. In fact, don’t look at numbers, at all. Numbers are meant to hide and deceive.
2. Think critically. Especially on stories around impact. Reach isn’t impact. Touch-points aren’t healthcare. Technology can’t solve problems that technology can’t solve. Innovation is a buzzword unless and until innovation leads to inclusion.
3. Be political. In thoughts, actions, and choices.
4. Be aware, call out, and discuss things like above with raw honesty. Reality is shaped by what we accept silently.