Question

我有一个3列的表格。我需要生成基于特定“诊断”执行的表的值（在“测试”列中）的组合（总是以3表示）。但是，一个特定的诊断有可能具有2个或更少的测试，在这种情况下，即使有2个值，逻辑仍将输出该组合。参考下表，对于每个cust_id，都有一个“诊断”列，基于该列执行“测试”。现在，对于每个诊断值组，我需要在“测试”列中生成相应值的唯一组合。请注意，组合应始终具有3个值（其中值> = 3），但是对于诊断而言，小于3个值（1或2）的情况下，仍应输出相应的组合（具有可用的1或2个值并替代空值代替不可用的值。

患者：

pat_id | diagnosis | tests

1001 | Thyroid | CAT
1001 | Thyroid | MRI
1001 | Thyroid | Blood

1001 | Tonsil  | CAT
1001 | Tonsil  | MRI
1001 | Tonsil  | Blood
1001 | Tonsil  | RAPID

1002 | Pneumonia | MRI
1002 | Pneumonia | Eliza

1003 | Bronchitis | X-Ray

因此，对于pat_id = '1001'和diagnosis = 'Thyroid'，我们看到“测试”具有3个不同的值。因此，只能有1个唯一的组合，即{CAT, MRI, Blood}。同样，对于pat_id = '1001' and diagnosis = 'Tonsil'，我们在“测试”列中看到4个不同的值。因此，将有4个组合，即{CAT, MRI, Blood}, {CAT, MRI, RAPID}, {MRI, Blood, RAPID} & {CAT, blood, RAPID}。对于pat_id = '1002'，仅存在两个唯一值。因此，组合将仅为1，即{MRI, Eliza}。与此类似，pat_id = '1003'仅具有1个值，即X射线，因此对于{X-Ray}，输出应为'1003'。

像这样，我需要为组中的所有诊断值生成相似的组合，最后，确定该表中出现最大次数的唯一组合。输出应为表中出现次数最多的组合。

到目前为止，下面的sql返回具有3个或更多值的所有组合。但是，它无法输出小于3的值。意思是，1002 & 1003的值小于3，因此没有输出，但是需要输出。解决方案还需要处理。

select p1.pat_id, p1.diagnosis, p1.tests, p2.tests, p3.tests
from patient p1 join
     patient p2
     on p1.pat_id = p2.pat_id and p1.diagnosis = p2.diagnosis and
        p1.tests < p2.tests join
     patient p3
     on p2.pat_id = p3.pat_id and p2.diagnosis = p3.diagnosis and
        p2.tests < p3.tests ;

还请说明如何确定哪种组合最多。谢谢。

Answer 1

您可以将查询与左联接一起使用，以允许第二个和第三个测试为NULL。但是随后您将需要删除具有两个或多个测试的组的带有NULL的行。您可以使用相关的（相关的）COUNT(*)子查询来实现：

select
  p1.pat_id,
  p1.diagnosis,
  p1.tests as test1,
  p2.tests as test2,
  p3.tests as test3
from patient p1
left join patient p2
  on  p2.diagnosis = p1.diagnosis
  and p2.pat_id = p1.pat_id
  and p2.tests > p1.tests
left join patient p3
  on  p3.diagnosis = p1.diagnosis
  and p3.pat_id = p1.pat_id
  and p3.tests > p2.tests
where
  case (
    select count(*)
    from patient p
    where p.diagnosis = p1.diagnosis
      and p.pat_id = p1.pat_id
  )
    when 1 then true
    when 2 then p2.tests is not null
    else        p3.tests is not null
  end
order by p1.pat_id, p1.diagnosis

结果：

| pat_id | diagnosis  | test1 | test2 | test3 |
| ------ | ---------- | ----- | ----- | ----- |
| 1001   | Thyroid    | Blood | CAT   | MRI   |
| 1001   | Tonsil     | CAT   | MRI   | RAPID |
| 1001   | Tonsil     | Blood | MRI   | RAPID |
| 1001   | Tonsil     | Blood | CAT   | MRI   |
| 1001   | Tonsil     | Blood | CAT   | RAPID |
| 1002   | Pneumonia  | Eliza | MRI   |       |
| 1003   | Bronchitis | X-Ray |       |       |

View on DB Fiddle

要按组合的出现顺序对不同的组合进行排序，只需将其修改为GROUP BY ... ORDER BY COUNT(*)查询：

select 
  p1.tests as test1,
  p2.tests as test2,
  p3.tests as test3,
  count(*) as cnt
from patient p1
left join patient p2
  on  p2.diagnosis = p1.diagnosis
  and p2.pat_id = p1.pat_id
  and p2.tests > p1.tests
left join patient p3
  on  p3.diagnosis = p1.diagnosis
  and p3.pat_id = p1.pat_id
  and p3.tests > p2.tests
where
  case (
    select count(*)
    from patient p
    where p.diagnosis = p1.diagnosis
      and p.pat_id = p1.pat_id
  )
    when 1 then true
    when 2 then p2.tests is not null
    else        p3.tests is not null
  end
group by p1.tests, p2.tests, p3.tests
order by cnt desc

结果：

| test1 | test2 | test3 | cnt |
| ----- | ----- | ----- | --- |
| Blood | CAT   | MRI   | 2   |
| CAT   | MRI   | RAPID | 1   |
| Blood | MRI   | RAPID | 1   |
| Eliza | MRI   |       | 1   |
| X-Ray |       |       | 1   |
| Blood | CAT   | RAPID | 1   |

View on DB Fiddle

Answer 2

我认为您想要left join和group by：

select p1.tests, p2.tests, p3.tests, count(*)
from patient p1 left join
     patient p2
     on p1.pat_id = p2.pat_id and p1.diagnosis = p2.diagnosis and
        p1.tests < p2.tests left join
     patient p3
     on p2.pat_id = p3.pat_id and p2.diagnosis = p3.diagnosis and
        p2.tests < p3.tests
group by p1.tests, p2.tests, p3.tests
order by count(*) desc;

我不确定diagnosis是否也应包含在结果集中。当您描述结果时，似乎并非如此，但对我来说是有道理的。

生成组合并确定哪个发生最多

2 个答案: